Headless CMS server device, AI-integrated server device, CMS server program, and AI-integrated server program

Abstract

This approach maintains the advantages of a headless CMS while enabling consistent content manipulation by an AI agent. [Solution] The headless CMS server device of the present invention includes a database that manages content IDs and associated attribute information in association. The AI ​​collaboration interface means receives content identification keywords for identifying the target of operation, which are extracted by the AI ​​agent from the user's instruction text. The ID identification means searches for information corresponding to the keyword in the attribute information in the database to identify the content ID.

Description

Technical Field

[0001] The present invention relates to a headless CMS server device, an AI-linked server device, a CMS server program, and an AI-linked server program.

Background Art

[0002] Conventionally, a CMS (Content Management System) for constructing, managing, and operating websites has been widely known. Usually, a CMS has a user interface (hereinafter referred to as an editing UI) for directly editing content, and even if editors in each management department do not have sufficient specialized knowledge, they can edit (update, add, etc.) web pages without going through web designers or engineers, so efficient website operation is possible under a division of labor system. In particular, since web pages can be added and updated based on templates (design prototypes) designed by web designers, there are few cases where there are significant deviations from the designs and user interfaces intended by the web designers for each page.

[0003] In recent years, headless CMSs have emerged that completely separate content management, editing, and distribution functions (backend) from the display screen (frontend), eliminating the need for a head or frontend (viewer) (see, for example, Patent Document 1). A headless CMS is a system specialized in managing, editing, and distributing content data, and displays websites by linking with a separately built frontend via an API. While headless CMSs are technically more difficult to implement than conventional CMSs, they offer superior versatility, security, and design flexibility. Because the frontend and backend are completely separated, system changes can be made without affecting each other, CMS functions can be added later, and there are advantages such as faster page display and reduced server load. Headless CMSs are also suitable for distributing data to various multi-channels with different frontends, such as apps, smart devices, and digital signage, in addition to websites.

[0004] On the other hand, the Model Context Protocol (MCP) has been proposed as a common standard for applications (such as AI models) to connect with external data sources and tools in a secure and standardized manner. MCP is a protocol that abstracts data exchange between different systems, and systems that adopt it can expose their system's functions and data as tools to AI agents that autonomously perform tasks based on user instructions.

[0005] As a result, AI agents can freely operate a wide range of information processing systems via MCP, such as business management tools, databases, and local file operations, and the integration of MCP into these systems is progressing (see, for example, Patent Document 2). Under these circumstances, even in the headless CMS mentioned above, by equipping it with a server equipped with MCP (MCP server), automated content manipulation by AI agents becomes possible. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2024-178099 [Patent Document 2] Japanese Patent Publication No. 2025-65338 [Overview of the project] [Problems that the invention aims to solve]

[0007] In conventional headless CMSs, when updating, retrieving, or deleting specific content (for example, a specific blog post), it is common practice to specify the target content using a unique identification number called a "content ID" that the system automatically assigns to each piece of content. Therefore, conventional headless CMSs had the following problems:

[0008] One issue is the burden on users. Due to its headless structure, which means it does not have its own display screen (frontend), the link between the "Content ID," a mechanically assigned identifier within the system with no descriptive meaning, and the "URL" and "article title" published on the actual website is difficult for users to intuitively understand. Therefore, when a user tries to update a specific article via an AI agent, the AI ​​agent / AI model does not have prior knowledge of the dynamic ID within the system. As a result, it must first investigate and identify the "Content ID" of the article in question via the management screen before inputting instructions to the AI, making the process extremely cumbersome. This hinders intuitive content management using AI and presents a high barrier to entry for users without specialized knowledge.

[0009] Another issue is the inherent constraints stemming from the system architecture. In non-headless CMSs where the database and frontend are integrated (such as WordPress), the system itself manages the display URLs, making it relatively easy to specify content by URL. In contrast, in headless CMSs where the frontend is completely externalized, the system (backend) does not directly know "which URL the content is displayed at on the actual site." Therefore, there was no convenient way for the user to directly retrieve the corresponding managed item (system ID) within the CMS from the information of the actual website (frontend) displayed in the browser, via an interactive interface.

[0010] This invention was proposed in view of the above points, and in one aspect aims to enable consistent content manipulation by an AI agent using intuitive information such as URLs and titles, without requiring awareness of system-specific identifiers, while maintaining the advantages of a headless CMS. [Means for solving the problem]

[0011] To solve the above problems, the headless CMS server device according to the present invention is a headless CMS server device configured to communicate with an AI agent device, and includes a database that manages each piece of content by linking a content ID, which is a system-specific identifier, with attribute information associated with the content, and comprises an AI collaboration interface means that communicates with the AI ​​agent device via a predetermined communication protocol, and an ID identification means that searches for attribute information corresponding to the content identification keyword from the attribute information stored in the database based on the content identification keyword received by the AI ​​collaboration interface means, and identifies the content ID associated with the attribute information, wherein the AI ​​collaboration interface means receives the content identification keyword extracted from a user instruction sentence that includes an instruction to identify the content to be operated on by an AI model used by the AI ​​agent device. [Effects of the Invention]

[0012] According to embodiments of the present invention, one aspect is that consistent content manipulation by an AI agent can be performed while maintaining the advantages of a headless CMS. [Brief explanation of the drawing]

[0013] [Figure 1] This diagram shows an example of a CMS network configuration. [Figure 2] This diagram shows an example of a CMS architecture configuration. [Figure 3] This diagram shows an example of a CMS server hardware configuration. [Figure 4] This diagram shows an example of the software configuration for a CMS server. [Figure 5] This figure shows an example of the MCP access token issuance screen. [Figure 6] This diagram shows the AI ​​agent screen when creating new content. [Figure 7] This diagram shows the AI ​​agent screen when creating new content. [Figure 8] This diagram shows the AI ​​agent screen when creating new content. [Figure 9] This figure shows the AI ​​agent screen used when creating new content according to this embodiment. [Figure 10] This diagram shows the AI ​​agent screen when creating new content. [Figure 11] This diagram shows the AI ​​agent screen when creating new content. [Figure 12] This diagram shows the content creation screen as seen from the frontend. [Figure 13] This diagram shows the management screen used when creating content. [Figure 14] This diagram shows the AI ​​agent screen when updating existing content. [Figure 15] This diagram shows the AI ​​agent screen when updating existing content. [Figure 16] It is a diagram showing the AI agent screen when existing content is updated. [Figure 17] It is a diagram showing the AI agent screen when existing content is updated. [Figure 18] It is a diagram showing an operation example when specifying an operation target using various content-specific keywords. [Figure 19] It is a sequence diagram showing the execution procedure for specifying and updating existing content according to this embodiment. [Figure 20] It is a diagram showing an example of the network configuration of the CMS.

Mode for Carrying Out the Invention

[0014] Embodiments of the present invention will be described in detail with reference to the drawings. In order to solve the above-described problems, the present invention incorporates an AI agent and an MCP interface unit capable of communicating with a predetermined protocol (such as MCP) in a headless CMS server, and executes the following processes.

[0015] · Associative storage of keywords at the time of generation In the process of creating new content, attribute information that is easy for humans to identify, such as "slug" and "title" input via the AI agent, is associated with a "content ID" automatically assigned by the system and stored in the database.

[0016] · Reverse lookup identification by attribute information In the AI agent, when an instruction to acquire, update, or delete existing content in a chat format (natural language) by the user is given, the headless CMS server equipped with the MCP interface unit searches the database in its own device using "content-specific keywords" such as URL, title, slug, etc. included in the instruction as a trigger, and automatically identifies the corresponding content ID.

[0017] · Inference by the AI model and instruction conversion: The AI ​​model extracts search parameters (the field name to be searched and its value, which is a content-specific keyword) from the user's instructions and converts them into a search instruction format that can be accepted by the headless CMS server via the MCP interface.

[0018] • Search execution using a headless CMS server: When the headless CMS server receives a search command at the MCP interface, it converts it into an executable command that can search the database within its own device, and identifies the content ID corresponding to the content identification keyword.

[0019] This allows users to identify the content they want to manipulate using intuitive information without having to input the content ID, even if they don't know the system's (headless CMS's) specific content ID, through a chat-style interaction via the AI ​​agent's interface. This enables them to complete operations such as content updates entirely within the AI ​​agent's interface. A detailed explanation follows.

[0020] <System Configuration> (Network configuration) Figure 1 shows an example of the network configuration of the CMS according to this embodiment. Figure 2 shows an example of the architectural configuration of the CMS according to this embodiment. The CMS (Content Management System) 100 according to this embodiment includes a CMS server 10, an AI agent (MCP client) 20, and an AI model 40, which are connected via a network 50.

[0021] The CMS server 10 according to this embodiment is a server device for a SaaS (Software as a Service) type content management system for building, managing, and operating a website. The CMS server 10 performs tasks such as user management, various system settings, web page content generation via an editing UI, content data management such as documents and images, and website management. The front end of the website is exposed to website visitors via a web server (not shown).

[0022] Furthermore, the CMS server 10 according to this embodiment is a headless CMS specialized in the management, editing, and distribution of content data, and therefore does not include a front-end on the CMS side. The front-end, which is built separately by engineers under the design supervision of a web designer, and the CMS server 10 cooperate via an API (Application Programming Interface) to display the website to visitors (decoupled architecture).

[0023] Furthermore, the CMS server 10 according to this embodiment, in addition to the function of providing a management screen via a conventional browser, etc., is equipped with an AI agent 20 and an MCP interface unit (MCP function unit) 101 for communication in accordance with MCP (Model Context Protocol). MCP is a communication standard developed for AI models 40 to access external data sources and tools in a secure and standardized manner. In other words, the MCP interface unit is a dedicated interface for exposing the functions and data of the CMS server 10 to the external AI agent 20 via MCP, and the MCP interface unit 101 can perform read and write operations on content data managed on the CMS in response to requests from the AI ​​agent (MCP client) 20.

[0024] The MCP interface unit 101 is configured as a functional unit that intervenes between the AI ​​agent 20 and the CMS server 10 and mediates communication between the two. Its main functions are to interpret operation instructions sent from the AI ​​agent 20 in an MCP-compliant format, convert them into commands that the CMS server 10 can process and transmit them, and to send the response back to the AI ​​agent 20 in the MCP standard format. Therefore, the MCP interface unit 101 may be built to be integrated into the CMS server 10, or it may be configured as a physically independent enclosure or a virtual (logical) MCP server 101-2 (Figure 20).

[0025] The MCP interface unit 101 can be any interface for communicating with the AI ​​agent 20 according to a predetermined communication protocol, and the communication protocol is not limited to MCP. In this case, the MCP interface unit 101 can also be called an AI collaboration interface unit that communicates with the AI ​​agent 20 according to a predetermined communication protocol.

[0026] The AI ​​agent (MCP client) 20 is an application program that acts as a bridge between the inference results from the AI ​​model 40 and operations on the CMS server 10, based on user instructions. The AI ​​agent 20 is configured as a desktop application program pre-installed on the user terminal 21.

[0027] The AI ​​agent 20, for example, is equipped with a chat-style user interface. When the user interactively inputs instructions (e.g., "Update the web content of XX"), it makes an inference request to the AI ​​model 40, which analyzes the chat-style (natural language) instructions input by the user and generates a logical operation procedure (task sequence) and execution parameters that the CMS server 10 should execute. The AI ​​agent 20 then sends the actual work commands to the MCP interface unit 101 of the CMS server 10 via MCP communication based on this operation procedure, and also receives the execution results and presents them to the user.

[0028] The user terminal 21 is implemented as, for example, a tablet device or a personal computer, and is an information terminal device used by engineers, web designers, and editors. The user terminal 21 also has an application program pre-installed as an AI agent 20, also called an MCP client. The user terminal 21 can communicate with the AI ​​model 40 and the CMS server 10 equipped with MCP functionality via the installed AI agent 20.

[0029] AI Model 40 is an external intelligent processing platform equipped with a Large-Scale Language Model (LLM). AI Model 40 analyzes chat-format text based on user instructions sent from AI Agent 20 and is responsible for generating draft logical procedures and content necessary to achieve the objective of those instructions, or draft operational parameters to be sent to CMS Server 10.

[0030] Furthermore, the AI ​​model 40 communicates bidirectionally only with the AI ​​agent 20 and does not have a direct communication channel with the CMS server 10. Therefore, the AI ​​model 40 does not directly know the address of the CMS server 10, and the intermediary AI agent 20 mediates the communication, thereby blocking unintentional access by external AI models 40 and ensuring the security and controllability of the entire system.

[0031] Network 50 is a communication network including wired and wireless networks. Network 50 includes, for example, the Internet, public telephone networks, Wi-Fi®, Bluetooth®, etc.

[0032] (Hardware configuration) Figure 3 shows an example of the hardware configuration of the CMS server according to this embodiment. The CMS server 10 includes a CPU (Central Processing Unit) 11, ROM (Read Only Memory) 12, RAM (Random Access Memory) 13, storage device 14, and communication device 15.

[0033] The CPU 11 executes various programs and performs calculations. The ROM 12 stores programs necessary for startup. The RAM 13 is a work area that temporarily stores processing performed by the CPU 11 and stores data. The storage device 14 stores and saves various data and programs. The communication device 15 communicates with other devices via the network 50.

[0034] (Software configuration) Figure 4 shows an example of the software configuration of the CMS server according to this embodiment. The CMS server 10 has as its main functional units an MCP interface unit 101, an ID identification unit 102, an execution control unit 103, and a storage unit 109.

[0035] The MCP interface unit (AI collaboration interface unit) 101 has the function of communicating with the AI ​​agent 20 via MCP communication. The MCP interface unit 101 also has the function of receiving content identification keywords extracted from the user's instruction text, which includes instructions for identifying the content to be operated on, by the AI ​​model 40 used by the AI ​​agent 20, and responding to the AI ​​agent 20 with the processing results.

[0036] The ID identification unit 102 has the function of searching for attribute information corresponding to the content identification keyword from the attribute information stored in the database based on the content identification keyword received by the MCP interface unit 101, and identifying the content ID associated with the attribute information.

[0037] The execution control unit 103 has the function of executing processing on the target content based on the identified content ID.

[0038] The memory unit 109 has the function of storing in the storage device 14 a database (DB) that manages each piece of content by linking it with a content ID, which is a system-specific identifier, and attribute information associated with that content.

[0039] It goes without saying that each of these functional units is a functional unit according to the present invention, and that the CMS server 10 also has other functional units related to various CMSs necessary for building, managing, and operating websites. Furthermore, each functional unit is realized by a program executed on the hardware resources such as the CPU, ROM, and RAM of the computer that constitutes the CMS server 10. These functional units may be read as "means," "module," "unit," or "circuit." In addition, the DB of the storage unit 109 in the CMS server 10 can also be placed on an external storage device on the network 50. Furthermore, the above-mentioned computer program and application program may be stored on a storage medium that is readable by the computer.

[0040] <Pre-configuration> First, the user performs the initial setup of the AI ​​agent 20 on their user terminal 21. The initial setup includes steps such as "installing the application," "obtaining authentication information (issuing a token)," "configuring the local configuration file," and "enabling delegated access rights," as shown below.

[0041] • Application installation The user pre-installs a desktop application on the user terminal 21 that functions as an AI agent 20. This application functions as an executable that sends inference requests to an external AI model 40 and controls MCP communication to the MCP interface unit 101 on the CMS server 10.

[0042] • Obtaining authentication information (token issuance) Figure 5 shows an example of the MCP access token issuance screen according to this embodiment. The user directly accesses the management screen (web interface, etc.) provided by the CMS server 10 using browsing software such as a browser, and requests the issuance of an "MCP access token," which is an authentication parameter specifically for linking with this system. The CMS server 10 generates an MCP access token consisting of a unique string based on the user's identification information and access rights, and outputs it to the user terminal 21.

[0043] An MCP access token is an authentication piece of information equivalent to an electronic master key, issued by the user to the AI ​​agent 20 to allow it to operate an external system (CMS server 10) on their behalf. Normally, a username and password are required to use the system, but directly entrusting these to the AI ​​model 40 or AI agent 20 poses a security risk. Therefore, this system uses an "MCP access token," which is a temporary key that grants only specific operational privileges.

[0044] This allows AI agent 20 to prove the user's legitimacy simply by presenting a token, without requiring the user to provide a password each time it communicates, enabling smooth automated operation (simplification and automation of identity verification). In the event that information in the configuration file is leaked externally, the user can immediately block unauthorized access to external systems without changing the main password by disabling only the relevant token from the management screen (limiting damage). The user can also limit the scope of work permitted to the AI ​​for each token, such as viewing-only keys or writing-only keys (permission control).

[0045] • Configuration of local configuration files Next, the user writes and saves the following information into a configuration file (for example, a configuration file in "json" format, etc.) specifically for the AI ​​agent 20 held on the user terminal 21: the identification information of the CMS server 10 to be connected to (server name, IP address, etc.), the execution command or path information for establishing communication with the MCP interface unit 101 (including command-line arguments using a package execution tool and information specifying dependent libraries required for execution), and the information of the MCP access token obtained above.

[0046] • Enable delegated access privileges The AI ​​agent 20 reads the above configuration file upon startup and stores the MCP access token written internally in memory. This enables the AI ​​agent 20 to automatically add the token to the header information of MCP communications, etc., based on the inference results obtained from the AI ​​model 40.

[0047] Through the above initial setup, a dynamic state is established in which the AI ​​agent 20 can access the CMS server 10 via the MCP interface unit 101 and manipulate the content data 100 on behalf of the user, following the operation procedures generated by the AI ​​model 40 after analyzing the user's instructions.

[0048] <Content creation example> First, the user (engineer) defines the structure of the content data for the web page in advance on the administration screen of the CMS server 10. In this embodiment, an example is shown of creating an announcement (model) to be posted on a corporate website (space) as content. The announcement model defined on the corporate website (space) has a configuration that includes the following fields (data items) which have been pre-modeled as content. • Article Title ·media • Article text (Content) • Meta-title • Meta description Slug • Category

[0049] (Example of content creation from the AI ​​agent screen) Figure 6-11 shows the AI ​​agent screen when creating new content (web page) according to this embodiment. The series of interactive processes by which the user builds a new web page via the AI ​​agent interface (chat screen) will be described below.

[0050] First, when the user enters "I want to create content" in the AI ​​agent's interface (chat screen), the AI ​​agent 20 forwards this information to the AI ​​model 40 (Figure 6). The AI ​​model 40 interprets the user's intention as "create new content" and determines that it should check the current configuration environment on the CMS server 10. The AI ​​model 40 generates an execution command to obtain information about the configuration environment and passes it to the AI ​​agent 20. Based on this command, the AI ​​agent 20 starts communicating with the CMS server 10 via the MCP interface unit 101.

[0051] The AI ​​agent 20 receives data returned from the CMS server 10 via the MCP interface unit 101 and feeds that data back to the AI ​​model 40. The AI ​​model 40 analyzes the raw data (JSON, etc.) such as "Available space: Corporate website (ID:xxxxx)", formats it into user-friendly natural language options, and displays them to the AI ​​agent 20 in a conversational format (Figure 7). It also displays a confirmation message, "Do you want to use this space to create content?", and if the user agrees with "Yes", the AI ​​agent 20 forwards the "Yes" information to the AI ​​model 40, which recognizes this as an agreement to proceed to the next step (Figure 7).

[0052] The AI ​​model 40 generates an execution command to the AI ​​agent 20 again to retrieve information from the CMS server 10 in order to identify the "model" defined within the confirmed space (corporate website), and passes this command to the AI ​​agent 20. Based on this command, the AI ​​agent 20 starts communicating with the CMS server 10 via the MCP interface unit 101. The AI ​​agent 20 receives the data returned from the CMS server 10 via the MCP interface unit 101 and feeds its contents back to the AI ​​model 40. The AI ​​model 40 formats the returned model information such as "Company Profile," "News," "Products / Services," and "Blog" into a user-friendly natural language format, such as numbered options, and displays it on the screen of the AI ​​agent 20 (Figure 8).

[0053] When the user selects "Notification" as the model, the AI ​​agent 20 transfers that information to the AI ​​model 40. The AI ​​model 40 then generates an execution command to the AI ​​agent 20 to obtain field information in order to identify what input fields are required for the selected "Notification" model, and passes this command to the AI ​​agent 20. Based on this command, the AI ​​agent 20 starts communicating with the CMS server 10 via the MCP interface unit 101 and obtains the definition information of the model from the CMS server 10 via the MCP interface unit 101.

[0054] The AI ​​model 40 analyzes the model definition information received via the AI ​​agent 20 in detail. The AI ​​model 40 identifies seven essential or recommended field items (article title, media, article body, meta title, meta description, slug, and category classification) for the structure of a web page, creates an input request message organized to make it easy for the user to sequentially enter each item, and passes it to the AI ​​agent 20. The AI ​​agent 20 displays this on the screen (Figure 9).

[0055] Once the user has entered all the field data, the AI ​​agent 20 transfers that information to the AI ​​model 40 (Figure 10). The AI ​​model 40 generates an execution command for "Create New Content" that includes all the field data and passes it to the AI ​​agent 20. Based on this command, the AI ​​agent 20 sends the data to the CMS server 10 via the MCP interface unit 101. After the CMS server 10 completes its content creation process, the AI ​​agent 20 feeds back the execution results (content ID, etc.) returned from the CMS server 10 via the MCP interface unit 101 to the AI ​​model 40. The AI ​​model 40 confirms that the content has been successfully generated and displays the AI ​​agent 20 with the message "Draft creation complete!" along with information about the created content, such as the issued content ID and the management screen URL including the user-specified slug (Figure 11).

[0056] Once the above content creation process is complete, the creation of the new content (web page) is finished. The published new content can be accessed, for example, from the front end via its URL on the actual website (Figure 12).

[0057] (supplement) On a webpage used for announcements, the "article title" and "article body" are the actual content, while the "meta title" and "meta description" are information for search engines. The "article title" is assigned by the user to best represent the "article body."

[0058] The entered field information is saved as content data on the CMS server 10, and a unique content ID (for example, "ID: abc12345") is automatically assigned by the CMS server 10. The content ID is an internal identifier used to uniquely manage the data of the content (web page).

[0059] • A "slug" is a unique identifier for a particular webpage and constitutes part of the URL (the end of the URL) when it is published on the actual website. Users can enter a string that is easy to identify (recognize).

[0060] The public URL is a link that provides direct access to the generated webpage. The end of this URL (the path portion) will reflect the "slug" string entered by the user.

[0061] A "category" is a tag used to classify web pages into common types based on their content. Categories can be created and used with strings that are easy for users to identify (recognize).

[0062] Typically, information such as "article title," "slug," and "category" is set using natural language or meaningful strings determined by the user, in order to make it easier for people to identify the content.

[0063] In another embodiment of this design, the AI ​​agent 20 (or an acquisition means such as a browser extension that works in conjunction with it) can be configured to automatically acquire a public URL from the user's browsing screen.

[0064] Figure 13 shows the management screen used when creating content (web pages) according to this embodiment. Needless to say, the CMS server 10 according to this embodiment can manage content not only using the AI ​​agent 20 described above, but also directly with users via a management screen accessible from a web browser or the like.

[0065] (Content identification and updating from the AI ​​agent screen) Figure 14-17 shows the AI ​​agent screen during the updating of existing content (web pages) according to this embodiment. When identifying and updating existing content (web pages), it is possible to identify the content to be operated on using keywords that a person intuitively understands and remembers, without directly specifying the "Content ID," which is a system-specific management number. This will be explained below.

[0066] First, when a user inputs an operation instruction such as "I want to update (modify) the content" into the AI ​​agent 20's interface, the AI ​​agent 20 forwards the instruction to the AI ​​model 40 (Figure 14). The AI ​​model 40 determines that a process is needed to uniquely identify the target of the operation, generates guidance to allow the user to select a method for identifying the target of the operation, and passes this guidance to the AI ​​agent 20.

[0067] AI Agent 20 displays several identification methods as options, such as "1. Specify by URL," "2. Specify by slug," "3. Specify by title," and "4. Other (keyword search, direct input of content ID, etc.)" (Figure 15). This allows users to start the update process in a way that makes it easy for them to identify the content they want to work with.

[0068] The user enters content-specific keywords to identify the content they wish to manipulate, following the guidance provided. These keywords can be selected from attribute information associated with the content (e.g., "URL (part of the slug)", "slug", "title", "other").

[0069] In general, when a user considers updating specific content (web pages), they examine the content of the page as it is publicly available on the website via a web browser, and consider areas for correction or improvement. Therefore, when identifying the content to be worked on, using the "public URL" displayed in the browser's address bar as the keyword is the most natural workflow for the user and makes it easy to obtain input information (copy and paste, etc.). Accordingly, in this embodiment, the user can preferentially input the website's public URL (e.g., https: / / www.xxx.jp / ja / ... / newproducts-2026) as the content identification keyword (Figure 15).

[0070] The entered URL information is sent to the AI ​​model 40 via the AI ​​agent 20. The AI ​​model 40 recognizes that the input string is a fully qualified domain name (FQDN) URL including the protocol (https) and domain name, and if a URL is entered, it performs an analysis process to extract the "slug (path portion)" contained at the end of the URL.

[0071] The AI ​​model 40 generates a search command (Tool Call) using the extracted slug and other information as parameters, and passes it to the AI ​​agent 20. The AI ​​agent 20 sends the search command to the CMS server 10 via the MCP interface unit 101.

[0072] The CMS server 10 performs a reverse lookup on the database that manages the content within its device, based on the received keyword (in this case, the "slug" portion). Specifically, it retrieves the "content ID," which is normally only referenced within the system, from human-related attribute information such as the "slug" and "title." If the search results identify the content ID corresponding to the keyword (in this case, the "slug"), the CMS server 10 returns the content ID and the current content information to the AI ​​agent 20 via the MCP interface unit 101. The AI ​​agent 20 then feeds back the search results (content ID and content information, etc.) returned from the CMS server 10 to the AI ​​model 40.

[0073] Once the AI ​​model 40 identifies the target content, it prompts the user to input specific changes, such as, "How would you like to update this content? Please tell us which fields (items) you want to update and what you want to update" (Figure 16).

[0074] When a user inputs instructions in natural language, such as "I want to update the article title to 'New Product Release in 2026 (Planned).' I also want to add a note to the end of the article body," the input instructions are sent from the AI ​​agent 20 to the AI ​​model 40, which analyzes these instructions. The AI ​​model 40 generates a content update execution command that links the update content (which fields, for example, Title and Content, should be rewritten and how) with the identified content ID, and passes this command to the AI ​​agent 20.

[0075] When the AI ​​agent 20 sends the generated content update execution command to the CMS server 10 via the MCP interface unit 101, the CMS server 10 actually executes the content update process.

[0076] After the content update is complete, the AI ​​agent 20 feeds back the execution results returned from the CMS server 10 to the AI ​​model 40. The AI ​​model 40 confirms that the content was generated successfully and generates a user-friendly response / message in natural language, such as "Content has been updated!", and displays the issued content ID and updated content information to the AI ​​agent 20 (Figure 17). The user can also check the links (URLs) after the update (correction) has been reflected, as well as the updated field information.

[0077] Figure 18 shows an example of how to identify the target content using various content-specific keywords. Users can identify the target content by entering the following attribute information individually or in combination, even if it is not in URL format.

[0078] When a user enters slug information (for example, "slug is newproducts-2026"), the AI ​​model 40 recognizes that this is a public path (slug), and the CMS server 10 performs a search on the slug field in the database to retrieve and identify the corresponding content ID (Figure 18(a)).

[0079] When a user enters title information (for example, "A page titled '2026 New Product Release List'"), the AI ​​model 40 recognizes the entered string as a title attribute, and the CMS server 10 searches the database using the title field as a key to retrieve and identify the corresponding content ID (Figure 18(b)).

[0080] When a user enters category classification information (for example, "Category classification is 'Announcements'"), the AI ​​model 40 recognizes the entered category classification as a tag, and the CMS server 10 searches the database using the category classification as a key to retrieve and identify the corresponding content ID (Figure 18(c)).

[0081] When a user enters free-word information (for example, "The page that lists the new product 'xxxx' from 2026."), the AI ​​model 40 extracts keyword values ​​such as "2026," "new product," and "xxxx," and generates a compound search command combining these multiple keywords. The CMS server 10 performs a compound search combining the multiple keywords and retrieves and identifies the corresponding content ID (Figure 18(d)).

[0082] Furthermore, if the CMS server 10 finds multiple content (web pages) that match the hit criteria in its search results, the following interactive filtering process may be executed. Cases where multiple content (web pages) match the hit criteria include, for example, multiple content with the same title, multiple content classified in the same category, or multiple content containing the same keyword. On the other hand, when based on a unique URL or slug, there is only one content (web page) that matches the hit criteria.

[0083] When the CMS server 10 returns multiple content IDs and content information, the AI ​​model 40 compares and organizes them, and presents a list of content candidates on the AI ​​agent 20's screen along with a message prompting the user to make a decision (for example, "xx matching content candidates were found. Which one would you like to update?"). When the user responds with, for example, a number or additional information from the presented content candidates, the AI ​​agent 20 transfers that selection information to the AI ​​model 40. Upon receiving this, the AI ​​model 40 uniquely identifies (confirms) the content ID to be operated on and proceeds with the content update process for that identified (confirmed) content.

[0084] In this way, in addition to unique URLs and slugs, attribute information linked to the content is used as content identification keywords, and when multiple hits occur, the target content is dynamically narrowed down through dialogue with the AI. As a result, users can identify the desired target content through human memory or specification / instruction via natural language chat input, without having to be aware of the content IDs within the CMS system, and can then perform the desired content operation on the CMS server 10.

[0085] Furthermore, since this attribute information is set by the user themselves when creating the content, or is information that users see on a daily basis on the website after it is published, it has the characteristic of being much easier for users to remember and identify intuitively compared to content IDs that are automatically generated by the system.

[0086] Furthermore, it is also possible to automatically obtain the public URL from browsing software such as a web browser, without requiring the user to manually enter the URL as described above. In this configuration, an extension program (hereinafter referred to as the "automatic acquisition tool") capable of obtaining information about the currently viewed tab is pre-installed in the web browser used by the user. When the user is viewing specific content (web page) in the first tab of the browser and inputs a natural language instruction such as "Update the content of this article" to the AI ​​agent 20 running in a second tab or side panel within the same browser, the automatic acquisition tool is triggered to extract the public URL (the extracted public URL is input to the AI ​​agent 20 in the same way as when the user manually enters the URL). With this configuration, the user is saved the trouble of explicitly specifying that the instruction "this article" refers to "the content corresponding to the currently viewed URL," enabling seamless interactive operation.

[0087] The mechanism by which the automated acquisition tool detects input operations involves the tool constantly monitoring the AI ​​agent 20's dialogue screen (DOM monitoring, etc.) and, the moment it detects that the user has sent a message, it acquires the public URL from the address bar of the first tab in the background. Alternatively, the AI ​​agent 20's screen may be configured to include a dedicated interface (such as a send button) for simultaneously sending instructions and acquiring the URL being viewed.

[0088] <Processing sequence for identifying and updating content> Figure 19 is a sequence diagram showing the procedure for identifying and updating existing content according to this embodiment. Each processing unit of the CMS server 10 and the user terminal 21 having the AI ​​agent 20 reads and executes a program capable of performing the processing, thereby realizing each step of the following sequence (hereinafter referred to as "S").

[0089] S1: First, the user inputs operation instructions and content-specific keywords (URL, slug, title, etc.) for the content they wish to operate on via the interactive interface on the AI ​​agent 20. The AI ​​agent 20 then transfers this natural language information to the AI ​​model 40 (Figures 14 and 15).

[0090] S2: AI model 40 analyzes the received information and extracts attribute information (content-specific keywords) specified by the user. Next, it generates a search command using those keywords as search parameters and returns it to AI agent 20.

[0091] S3: Based on the search command, the AI ​​agent 20 sends a search request containing content-specific keywords to the CMS server 10 (MCP interface unit 101) in accordance with the MCP.

[0092] S4: The CMS server 10 scans its internal database using the content identification keyword received via the MCP interface unit 101 as a key. At this time, the CMS server 10 retrieves a system-specific content ID associated with the attribute information (reverse lookup).

[0093] S5: The CMS server 10 returns the identified content ID and the current content associated with that ID (current title, body text, URL, various settings, etc.) to the AI ​​model 40 via the AI ​​agent 20. This allows the AI ​​model 40 to uniquely recognize which data (record) within the system the ambiguous keyword provided by the user refers to, using the same content ID as the CMS server 10.

[0094] S6: Based on the returned identified content ID and the current content associated with that ID, the AI ​​model 40 generates a message to report the identification result to the user (for example, a confirmation message such as "We have identified this content...") and passes it to the AI ​​agent 20.

[0095] S7: Based on this generated message, AI agent 20 displays the identified content information on the screen (Figure 16). This allows the user to reliably confirm, through interaction with the AI ​​model, that the web page they intended to access has been correctly identified.

[0096] S8: When the user confirms the display of identified content information on the screen and enters specific modification instructions for that identified content (for example, "Change title to ~"), the instructions are forwarded to AI model 40.

[0097] S9: AI model 40 generates an update execution command that combines the already determined content ID with the modification details based on user instructions, and passes this to AI agent 20.

[0098] S10: The AI ​​agent 20 receives an update execution command and, in accordance with the MCP, sends an update request specifying the content ID to the CMS server 10 (MCP interface unit 101).

[0099] S11: The CMS server 10 updates the database record corresponding to the specified content ID and sends a response to the AI ​​agent 20 indicating that the process has been completed successfully. When the AI ​​agent 20 receives the processing completion response (updated data, etc.) from the CMS server 10, it sends it to the AI ​​model 40.

[0100] S12: After confirming that the update has been completed successfully, AI model 40 generates a message for the user summarizing "what corrections were made" and prompting them to check the corrected parts, and passes this to AI agent 20.

[0101] S13: Based on this generated message, AI agent 20 notifies the user that the series of tasks are complete by displaying a notification that "corrections have been made," as well as link information such as the updated article title and the URL for publication on the screen (Figure 17).

[0102] <Examples of Use> This invention is also useful for content manipulation initiated from external services and various interfaces. Several specific use cases are illustrated below. (Case 1) This case demonstrates how the AI ​​agent 20 integrates statistical data obtained from external access analysis tools (such as Google Analytics 4) with content information within the CMS server 10 to achieve content optimization based on external analysis data.

[0103] When a user gives instructions from the AI ​​agent 20 screen, such as "Identify articles with high access numbers but low conversion rates (CVR)," the AI ​​model 40 interprets these instructions and generates execution commands to obtain the necessary data from the analysis tool. Upon receiving these execution commands, the AI ​​agent 20 accesses an external analysis server via MCP based on pre-configured authentication information and obtains statistical data.

[0104] AI Model 40 also analyzes the acquired statistical data and identifies "problematic articles" that meet the criteria. The AI ​​Agent 20 screen displays a list of the titles and public URLs of these articles. Furthermore, AI Model 40 reads the content of the target articles, analyzes the causes of the problems, and then proposes specific article revisions to improve the conversion rate (CVR) to the user from the AI ​​Agent 20 screen.

[0105] When the user accepts the proposal, the AI ​​model 40 adopts the public URL on the list as a content identification keyword and generates an update execution command combining the URL with the specific modification details. Based on this command, the AI ​​agent 20 sends a content update request to the CMS server 10 via the MCP interface unit 101. The CMS server 10 scans the database based on the received URL, identifies the corresponding system-specific content ID by reverse lookup, and then executes the update of the specific article content for that ID, completing the process.

[0106] In this case, the previously separate "statistical information (URL) on the access analysis tool" and "management data (ID) on the CMS" are dynamically combined via an AI agent. Users are freed from the manual task of switching between the analysis screen and the management screen to identify and search for target articles, and can seamlessly complete everything from "data analysis" to "updating actual data" through natural language interaction alone. This significantly reduces the workload involved in website improvement and operation, and structurally eliminates human errors such as misidentifying content IDs.

[0107] (Case 2) This is an example of instantly reflecting topics from social networking services (SNS) on relevant pages of a company's website. A user presents a link to a specific SNS post to the AI ​​agent 20's chat interface and instructs it to "incorporate this content and update the content of the related product introduction page (specified URL)." The AI ​​model 40 identifies the content ID from the provided URL and quickly updates the official website by reflecting the content of the SNS post (for example, as a customer testimonial).

[0108] (Case 3) This is an example of completing an urgent correction instruction from a mobile user terminal 21 via a business chat tool while away from the office. The user instructs the AI ​​agent 20's chat interface in natural language, saying, "Please correct the typographical error in the published article (specified by URL or slug)." Even in a mobile device environment where logging into the management screen or performing complex search operations is difficult, accurate corrections can be made using only information that is intuitive to humans (such as URLs).

[0109] (Case 4) This is an example of extending updates to the Japanese version of an article to its corresponding articles in other languages ​​(such as English). The user instructs, "I've updated the Japanese version (... / ja / article-01), so please reflect the changes in the English version of the article that has the same slug (common attribute information)." Even if the system's content IDs differ for each project or language, batch synchronization can be achieved by using the common "slug" as a key to reverse look up and identify the ID of each language project.

[0110] <Summary> In conventional headless CMSs, due to their structure, the impersonal content IDs within the system are not directly linked to the actual information on the website (URLs or article titles) (or are not mapped within the CMS). This resulted in the cumbersome task of users having to investigate and identify the content IDs in advance through the administration screen or other means, even when operating via an AI agent.

[0111] In this invention, when new content is generated, attribute information such as slugs and titles, which are easily identifiable and rememberable by humans, is linked to the content ID and stored. When operating on existing content, the user inputs attribute information such as slugs and titles in a chat format to identify the content they wish to operate on, and the AI ​​agent utilizes these as content identification keywords. The AI ​​model 40 extracts search parameters from the user's input instructions and converts them into commands, and the CMS server 10 equipped with the MCP interface unit 101 searches its own database to automatically identify the corresponding content ID.

[0112] This allows users to consistently identify the target of their actions and complete operations such as updating content using only intuitive information through a chat-style interaction with an AI agent, even without knowing the system's unique content ID.

[0113] In other words, according to this embodiment, while maintaining the advantages of a headless CMS, consistent content manipulation by an AI agent is possible using intuitive information such as URLs and titles, without having to be aware of system-specific identifiers.

[0114] Although the present invention has been described with specific examples in the form of preferred embodiments, it is clear that various modifications and changes can be made to these examples without departing from the broad spirit and scope of the invention as defined in the claims. In other words, the details of the examples and the accompanying drawings should not be construed as limiting the present invention.

[0115] (term) • Spaces: A space is a unit for managing site elements such as content, media files, models, and fieldsets. A model is a template for content. It is defined by combining fields. There are two types of models: a multi-type model that allows for the creation of an unlimited amount of content, and a single-type model that allows for the creation of only one piece of content. Since the layout of a web page is defined by its content elements, it is also called a layout model. Fields are a feature that allows you to add input forms to the administration screen. You can design content by combining multiple types of fields. Content is tangible data that embodies the model, specifically referring to management units such as web pages, blog posts, or categories. It is generated by inputting specific content into each field defined in the model, and is the primary subject of reverse lookup identification in this invention. • Media: Files uploaded by users to be incorporated into content. This can handle all types of files, including images, videos, and PDFs. Furthermore, regarding headless CMS, you can refer to "Patent Document 1" (Japanese Patent Publication No. 2024-178099) by the present applicant, etc.

[0116] (Note 1) A headless CMS server device (CMS server 10, Figure 1) is configured to communicate with an AI agent device (AI agent 20 is an installed user terminal 21) and has a database that manages each piece of content by linking it with a system-specific identifier, the content ID (e.g., "ID: abc12345") and attribute information associated with that content (e.g., "URL (part of the slug)", "slug", "title", "other"). An AI collaboration interface means (MCP interface unit 101, S3, S10 in Figure 19) communicates with the AI ​​agent device via a predetermined communication protocol (e.g., MCP), Based on the content identification keyword received by the AI ​​linkage interface means, an ID identification means (ID identification unit 102, S4 in Figures 16 and 19) searches for attribute information corresponding to the content identification keyword from the attribute information stored in the database and identifies the content ID associated with the attribute information, It has, The AI ​​collaboration interface means receives the content identification keyword (Figure 15) extracted from the user's instruction text (chat input text) which includes instructions for identifying the content to be operated on, using the AI ​​model (AI model 40:LLM) used by the AI ​​agent device. A headless CMS server device characterized by the following:

[0117] (Note 2) Based on the identified content ID, an execution control means (execution control unit 103, S5 and S11 in Figures 17 and 19) executes processing on the target content, It has, The AI ​​collaboration interface means responds to the AI ​​agent device with the result of the processing (Figure 17). A headless CMS server device as described in Appendix 1, characterized by the above. [Explanation of Symbols]

[0118] 10 CMS Servers 20 AI Agents 40 AI Models 50 Networks 100 CMS 101 MCP Interface Section 102 ID identification part 103 Execution Control Unit 109 Storage section

Claims

1. A headless CMS server device is configured to communicate with an AI agent device and includes a database that manages each piece of content by linking a content ID, which is a system-specific identifier, with attribute information associated with that content, An AI collaboration interface means that communicates with the AI ​​agent device via a predetermined communication protocol, Based on the content identification keyword received by the AI ​​linkage interface means, an ID identification means searches for attribute information corresponding to the content identification keyword among the attribute information stored in the database and identifies the content ID associated with the attribute information, It has, The AI ​​collaboration interface means receives the content identification keyword extracted from the user's instruction text, which includes instructions for identifying the content to be operated on, by the AI ​​model used by the AI ​​agent device. A headless CMS server device characterized by the following:

2. An execution control means that performs processing on the target content based on the identified content ID, It has, The AI ​​collaboration interface means responds to the AI ​​agent device with the result of the processing. A headless CMS server device according to claim 1, characterized by the following:

3. The aforementioned attribute information includes a slug that constitutes part of the public URL for each piece of content. The ID identification means identifies the content ID based on the slug received as the content identification keyword. A headless CMS server device according to claim 1, characterized by the following:

4. The aforementioned attribute information includes at least one of the strings contained in the title or body of the content, The ID identification means identifies the content ID based on information corresponding to at least a part of the string contained in the title or body received as the content identification keyword. A headless CMS server device according to claim 1, characterized by the following:

5. The AI-linked interface means receives the slug, which is extracted from a part of the URL of the screen information being viewed in the web browser, as the content identification keyword. A headless CMS server device according to claim 3, characterized by the above.

6. An AI collaboration server device is interposed between an AI agent device and a CMS server device equipped with a database that manages each piece of content by linking a content ID, which is a system-specific identifier, with attribute information associated with that content, and is configured to communicate with the AI ​​agent device, An AI collaboration interface means that communicates with the AI ​​agent device via a predetermined communication protocol, Based on the content identification keyword received by the AI ​​linkage interface means, an ID identification means searches for attribute information corresponding to the content identification keyword among the attribute information stored in the database of the CMS server device, and identifies the content ID associated with the attribute information. It has, The AI ​​collaboration interface means receives the content identification keyword extracted from the user's instruction text, which includes instructions for identifying the content to be operated on, by the AI ​​model used by the AI ​​agent device. An AI-linked server device characterized by the following:

7. A computer, which is a headless CMS server device configured to communicate with an AI agent device, has a database that manages each piece of content by linking a content ID, which is a system-specific identifier, with attribute information associated with that content, An AI collaboration interface means that communicates with the AI ​​agent device via a predetermined communication protocol, Based on the content identification keyword received by the AI ​​linkage interface means, an ID identification means searches for attribute information corresponding to the content identification keyword among the attribute information stored in the database and identifies the content ID associated with the attribute information, and make it work The AI ​​collaboration interface means is a CMS server program that receives content identification keywords extracted from a user's instruction text, which includes instructions for identifying the content to be operated on, by an AI model used by the AI ​​agent device.

8. Interposed between the AI ​​agent device and the CMS server device, which has a database that manages each piece of content by linking a content ID, a system-specific identifier, with attribute information associated with that content, is a computer that is an AI collaboration server device configured to communicate with the AI ​​agent device. An AI collaboration interface means that communicates with the AI ​​agent device via a predetermined communication protocol, Based on the content identification keyword received by the AI ​​linkage interface means, an ID identification means searches for attribute information corresponding to the content identification keyword among the attribute information stored in the database of the CMS server device, and identifies the content ID associated with the attribute information. and make it work The AI ​​collaboration interface means is an AI collaboration server program that receives content identification keywords extracted from a user's instruction text, which includes instructions for identifying the content to be operated on, by an AI model used by the AI ​​agent device.

Images