MCP protocol-based IFC model cognitive agent and working method thereof

Abstract

The application relates to an IFC model cognitive agent based on an MCP protocol and a working method thereof. The IFC model cognitive agent obtains and analyzes a processing requirement of an IFC model file proposed by a user in natural language through an intelligent hub, calls a filesystem-Server module to obtain a target IFC model file from a file system according to the user requirement, calls an IFC-MCP-Server module to query and analyze the target IFC model file, modify data and / or automatically process, completes automatic processing of the target IFC model file, and finally feeds back the processed target IFC model file to the user in a required format. In the overall operation process of the IFC model cognitive agent, an IFC-Viewer module is used to load and display the target IFC model file, a GLTF model file and a GLB model file, so that the user can check and review.

Description

Technical Field

[0001] This application relates to the field of IFC model file processing technology, and in particular to an IFC model cognitive agent based on the MCP protocol and its working method. Background Technology

[0002] Currently, conventional methods for utilizing and mining the value of BIM models based on Industry Foundation Classes (IFC) standards primarily rely on specialized BIM software platforms. User operation is often limited by the specific interfaces and complex operational logic of these platforms, requiring users to not only possess deep professional knowledge but also a thorough understanding of the specific software's functional structure and the data organization methods of the IFC standard. This operation mode, heavily reliant on graphical user interfaces for point selection, filtering, and attribute queries, suffers from significant bottlenecks when facing demands for rapid location, multi-dimensional correlation analysis, or batch processing of complex model information. These bottlenecks include lengthy and cumbersome operation paths, high cognitive load, steep learning curves, and low information acquisition efficiency. This severely limits the breadth of participation for non-professional users and the efficiency of professional users in deeply mining the value of model data, hindering the further release of the potential of BIM technology.

[0003] Meanwhile, large language models demonstrate powerful natural language understanding and generation capabilities, knowledge reasoning, and complex task decomposition abilities, providing a new paradigm for solving the challenges of information overload and complex system interactions. How to efficiently and intelligently process and understand the massive amounts of structured and semi-structured data contained in building models, and achieve seamless dialogic interaction between "human-model-data," has become a key challenge in improving the effectiveness of BIM applications and expanding their application scenarios. Summary of the Invention

[0004] Therefore, it is necessary to provide an IFC model cognitive agent based on the MCP protocol, including:

[0005] Intelligent Hub, IFC-MCP-Server Module, IFC-Viewer Module;

[0006] The intelligent hub includes a conversational AI framework built on a large language model, which is used to understand the user's processing requirements for IFC model files proposed in natural language by combining context, and to call the filesystem-Server module and IFC-MCP-Server module to create a workflow according to the user's requirements; and to add and apply the IFC-MCP-Server module and IFC-Viewer module to the intelligent hub.

[0007] The filesystem-Server module includes an open-source filesystem service, which is used to retrieve the target IFC model file from the file system according to the calling command;

[0008] The IFC-MCP-Server module is built on the FastMCP high-performance framework and the ifcopenshell underlying computing engine. It is used to query and analyze, modify data, and / or automatically process target IFC model files according to the calling instructions.

[0009] The IFC-Viewer module is built on a modern web technology stack based on Vue3 and Three.js, and is used to load and display target IFC model files, GLTF model files, and GLB model files.

[0010] Preferably, the IFC-MCP-Server module includes a query analysis module, a data modification module, and an automatic processing module;

[0011] The query and analysis module is used to query the geometric properties, engineering properties, containment system, and type hierarchy of target components in the target IFC model file;

[0012] The data modification module is used to modify the cross-model operation scenarios of target components in the target IFC model file, batch update the attributes of target components in the target IFC model file, and delete components in the target IFC model file;

[0013] The automatic processing module is used to perform integrated geometric and semantic transformation of the target IFC model file, as well as structured export.

[0014] Preferably, in the query analysis module, querying geometric attributes includes querying the coordinates and spatial location of the component; querying engineering attributes includes querying the name, information, material composition, fire resistance rating, and load-bearing characteristics of the component; querying the inclusion system includes the space where the component is located and all components within that space; querying the type hierarchy system includes querying all components corresponding to the category, querying components made of concrete, and querying the type of each component.

[0015] Preferably, in the data modification module, the cross-model operation scenario of modifying the target component in the target IFC model file includes copying the component in the target IFC model file to other IFC model files; the batch update of the attributes of the target component in the target IFC model file includes batch standardizing the names of the target components in the target IFC model file.

[0016] Preferably, in the automatic processing module, the geometric and semantic integration conversion of the target IFC model file includes exporting the target IFC model file as a GLB model file; the structured export of the target IFC model file includes exporting the target components in the target IFC model file as a CSV model file.

[0017] Preferably, the workflow of the IFC-MCP-Server module includes:

[0018] Tool method design includes designing tool methods for query analysis, data modification, and automatic processing. Each tool method includes a method name, input parameters, return value data type, text comments of statements in the method, and method body; the input parameters are the calling instructions.

[0019] Creating an MCP-Server instance includes constructing the MCP-Server instance by calling the FastMCP method in the mcp.server.fastmcp library;

[0020] Registering utility methods includes adding "@mcp.tool()" before the utility method or registering the utility method to the MCP-Server instance using "mcp.add(utility method name)";

[0021] Run the service, including running an MCP-Server instance and selecting the underlying transport method, which may be stdio or SSE protocol.

[0022] Preferably, the IFC-Viewer module includes a local model loading module, a GLTF or GLB model file loading module, and a component attribute acquisition module;

[0023] The local model loading module is used to obtain the target IFC model file in the local folder through the webkitdirectory property, and use the URL.createObjectURL() method to create a temporary access URL, which is then passed to the IFCLoader loader. The target IFC model file is then loaded and displayed based on the IFCLoader loader.

[0024] A GLTF or GLB model file loading module, used to load and display GLTF or GLB model files using the GLTFLoader loader in Three.js;

[0025] The component information acquisition module is used to obtain the vertex indices of the triangles on the target component detected by the mouse ray in the target IFC model file. Based on the vertex indices of the triangles, it calls the faceIndex property of the triangle to get the index of the triangle, and multiplies the triangle index by 3 to get the index of the first vertex of the triangle. It uses the geometry.attributes.expressID.array[vertexIndex] method to get the ExpressID of the target component, and uses the ifcLoader.ifcManager.getItemProperties() method to get the name, type, GlobalID, label, and geometric dimensions of the target component. Finally, it loads and displays the name, type, GlobalID, ExpressID, label, and geometric dimensions of the target component using the IFCLoader loader.

[0026] Preferably, the system also includes settings to optimize the performance of the component information acquisition module, including:

[0027] The geometry.toNonIndexed() method is used to convert indexed geometry on the target component into non-indexed geometry to simplify the construction of the boundary volume hierarchy.

[0028] The normal vectors of the vertices of the triangle face are recalculated using the geometry.computeVertexNormals() method to fix the lighting display problem caused by the geometry transformation;

[0029] The geometry.boundsTree = new MeshBVH(geometry) method is used to construct a boundary volume hierarchy for the geometry in order to reduce the complexity of ray detection.

[0030] Preferred conversational AI frameworks include Cherry-Studio, Claude, Fast-agent, and Dify.

[0031] On the other hand, a working method for an IFC model cognitive agent based on the MCP protocol is also provided, including:

[0032] Step 1: Obtain the processing requirements of the IFC model file submitted by the user in natural language;

[0033] Step 2: Use a large language model to analyze the processing requirements based on context and confirm the processing method for the target IFC model file;

[0034] Step 3: Based on the analysis process, call the filesystem-Server module to obtain the target IFC model file from the file system, and then call the IFC-MCP-Server module to perform query analysis, data modification, and / or automatic processing on the target IFC model file;

[0035] Step 4: Feed back the processed target IFC model file to the user in the required format; and during the overall operation of the IFC model cognitive agent, load and display the target IFC model file, GLTF model file, and GLB model file through the IFC-Viewer module so that the user can compare and review them.

[0036] Beneficial Effects: This IFC model cognitive agent acquires and analyzes the processing requirements of IFC model files submitted by users in natural language through an intelligent hub. Based on user requirements, it calls the filesystem-Server module to retrieve the target IFC model file from the file system, and the IFC-MCP-Server module to query, analyze, modify, and / or automatically process the target IFC model file, completing the automated processing of the target IFC model file. Finally, the processed target IFC model file is returned to the user in the required format. Furthermore, during the overall operation of the IFC model cognitive agent, the IFC-Viewer module loads and displays the target IFC model file, GLTF model file, and GLB model file for user comparison and review. This IFC model cognitive agent significantly improves the accessibility, parsing efficiency, and ease of operation of Building Information Modeling (BIM) data, lowers the application threshold, and unlocks data value, providing a core driving engine for the intelligent upgrading of the construction industry. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the structure of the IFC model cognitive agent based on the MCP protocol in the embodiments of this application.

[0039] Figure 2 This is a design example diagram of the tool method in the IFC-MCP-Server in the embodiments of this application.

[0040] Figure 3 This is a schematic diagram illustrating the effect of reading the IFC model file in an embodiment of this application.

[0041] Figure 4 This is a schematic diagram of the component properties obtained in the embodiments of this application.

[0042] Figure 5 This is a flowchart illustrating the working method of the IFC model cognitive agent based on the MCP protocol in this application embodiment. Detailed Implementation

[0043] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0044] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0045] like Figure 1 As shown, this embodiment provides an IFC model cognitive agent based on the MCP protocol, including:

[0046] Intelligent Hub, IFC-MCP-Server Module, IFC-Viewer Module;

[0047] The intelligent hub includes a conversational AI framework built on a large language model, which is used to understand the user's processing requirements for IFC model files proposed in natural language by combining context, and to call the filesystem-Server module and IFC-MCP-Server module to create workflow according to user requirements; and to add and apply the IFC-MCP-Server module and IFC-Viewer module to the intelligent hub.

[0048] In this embodiment, the conversational AI framework includes Cherry-Studio, Claude, Fast-agent, and Dify.

[0049] The filesystem-Server module includes an open-source filesystem service (such as the filesystem service built into the Cherry-Studio platform), which is used to retrieve the target IFC model file from the file system according to the call command. The filesystem service is mainly used to operate the file system that stores the IFC model file.

[0050] The IFC-MCP-Server module is the core of semantic operations around IFC model files. Based on the FastMCP high-performance framework and the ifcopenshell underlying computing engine, it is a protocol service layer that supports complex engineering logic. It fully implements three types of service modules: query analysis, data modification, and automatic processing, breaking through the mechanism of accurate conversion of natural language commands to model operations. It is used to perform query analysis, data modification, and / or automatic processing on the target IFC model file according to the calling command.

[0051] Specifically, the IFC-MCP-Server module includes a query analysis module, a data modification module, and an automatic processing module;

[0052] The query and analysis module is used to query the geometric properties, engineering properties, containment system, and type hierarchy of target components in the target IFC model file;

[0053] In the query analysis module, querying geometric attributes includes querying the coordinates and spatial location of components; querying engineering attributes includes querying the name, information, material composition, fire resistance rating, and load-bearing characteristics of components; the query inclusion system includes querying the space where the component is located and all components within that space; the query type hierarchy system includes querying all components corresponding to a category, querying components made of concrete, and querying the type of each component.

[0054] The query analysis module is designed with semantic parsing capabilities that cover multi-dimensional information, breaking through the limitations of traditional ID retrieval. It enables composite semantic queries based on natural language descriptions, such as "extraction of concrete staircase components" and "location of electromechanical equipment in the core tube". It supports dynamic association analysis and reverse relationship backtracking, such as locating the space to which the equipment belongs.

[0055] The data modification module is used to modify the cross-model operation scenarios of target components in the target IFC model file, batch update the attributes of target components in the target IFC model file, and delete components in the target IFC model file;

[0056] In the data modification module, cross-model operation scenarios for modifying target components in the target IFC model file include copying components from the target IFC model file to other IFC model files; batch updating the attributes of target components in the target IFC model file includes batch standardizing the names of target components in the target IFC model file.

[0057] The data modification module develops a component-level non-destructive editing structure, supporting cross-model operation scenarios such as component copying and migration, batch updates of attributes such as name standardization, and deletion of key components, while simultaneously maintaining a model topology inhibition verification mechanism.

[0058] The automatic processing module is used to perform integrated geometric and semantic transformation of the target IFC model file, as well as structured export;

[0059] In the automatic processing module, the geometric and semantic integration transformation of the target IFC model file includes exporting the target IFC model file as a GLB model file; the structured export of the target IFC model file includes exporting the target components in the target IFC model file as a CSV model file.

[0060] The automatic processing module can meet the needs of mobile collaboration and data analysis by building a data output channel for lightweight applications.

[0061] Furthermore, the workflow of the IFC-MCP-Server module includes:

[0062] Tool method design includes designing tool methods for query analysis, data modification, and automatic processing. Each tool method includes a method name, input parameters, return value data type, text comments of statements in the method, and method body; the input parameters are the calling instructions.

[0063] Tool method design differs from traditional program development design. In terms of design principles, it prioritizes broad applicability and high fault tolerance. This means that method input parameters are not directly input, but rather passed from upstream nodes in the workflow, built by the large language model in conjunction with the context and method description. Therefore, the same type of operation on different objects should be designed as a single method, with the operation methods for the input parameters covering all objects of that type. The broad applicability principle allows for fewer tokens to be consumed when the large language model is called; furthermore, the method establishes fault tolerance mechanisms through "try" statements, ensuring that the workflow is not interrupted by the failure of a local method call.

[0064] For example, to satisfy the requirement of retrieving elements belonging to categories such as "door", "window" or "wall", a method for "retrieving all elements of the same category" is designed. Based on the different input parameters, the target category is identified, thereby retrieving the expected elements, avoiding the need to design methods for each category unit.

[0065] Similarly, to help the large language model better understand the purpose, usage rules, and input / output of tool methods, the design should include the method name, input parameters (optional), return value data type, textual comments within the method, and the method body. Textual comments within the method refer to the explanatory text added in the form of comments after the method declaration and in the middle of the method body. These comments mainly include concise and clear explanations of the method's function and purpose, return value content and format, and, if necessary, supplementary logical relationships to supplement the code's execution mechanism and data mapping rules. Additionally, a "Requirements and Explanations for Input Parameter Acquisition" section should be included to prompt the large language model to lock the source of input parameters. Although the computer does not execute the comments when running the program, the large language model will interpret them when calling MCP services (including the filesystem-Server module and the IFC-MCP-Server module) to facilitate a deeper understanding of the method's function and rules. An example of a tool method design is shown below. Figure 2 As shown.

[0066] Create an MCP-Server instance, including constructing the MCP-Server instance by calling the FastMCP method in the mcp.server.fastmcp library.

[0067] Registering utility methods includes adding "@mcp.tool()" before the utility method or registering the utility method to the MCP-Server instance using "mcp.add(utility method name)".

[0068] Running the service involves running an MCP-Server instance and selecting the underlying transport method, which can be either stdio or the SSE protocol. The standard input / output method (stdio) starts the server as a child process, exchanging JSON-RPC messages with the client's parent process via stdin and stdout, suitable for local scenarios. The SSE protocol (Server-Sent Events) runs the server as an HTTP service, with clients communicating via SSE connections, suitable for remote or network scenarios. In this embodiment, the stdio transport method is used during local development and personal use, while the SSE transport method is used during the promotion and use phase.

[0069] The IFC model cognitive agent provided in this implementation breaks through the limitations of dynamically parsing natural language instructions into MCP service invocation instructions through a semantic mapping engine. A multi-level prompt word constraint framework is designed for LLM (Large Language Model) output, significantly improving the accuracy of attribute querying and data analysis of IFC model files, as well as the robustness of service invocation.

[0070] The IFC-Viewer module is built on a modern web technology stack of Vue3 and Three.js, and is used to load and display target IFC model files, GLTF model files, and GLB model files. The core objectives of the IFC-Viewer module include selecting an IFC folder, adding the names of all IFC models in the folder to a drop-down list, selecting a model, loading and displaying the model file, and then selecting a local GLTF or GLB model via a separate button and loading and displaying it.

[0071] Specifically, the IFC-Viewer module includes a local model loading module, a GLTF or GLB model file loading module, and a component attribute acquisition module;

[0072] The local model loading module retrieves the target IFC model file from the local folder via the webkitdirectory property, creates a temporary access URL using the URL.createObjectURL() method, passes the temporary access URL to the IFCLoader loader, and loads and displays the target IFC model file based on the IFCLoader loader. The reading effect of the target IFC model file is as follows. Figure 3 As shown;

[0073] In this embodiment, to avoid displaying the "IFCSPACE" spatial element during model loading and affecting the visual effect, the "setupOptionalCategories" method is used to set the "IFCSPACE" spatial element to be unselected. Specifically:

[0074] ifcLoader.ifcManager.parser.setupOptionalCategories({

[0075] [IFCSPACE]: false,

[0076] }).

[0077] A GLTF or GLB model file loading module, used to load and display GLTF or GLB model files using the GLTFLoader loader in Three.js;

[0078] The component information acquisition module is used to obtain the vertex indices of the triangles on the target component detected by the mouse ray in the target IFC model file. Based on the vertex indices, it calls the `faceIndex` property of the triangle to obtain the triangle's index, and multiplies the triangle's index by 3 to get the index of the first vertex of the triangle. It then uses the `geometry.attributes.expressID.array[vertexIndex]` method to obtain the target component's ExpressID. Based on the ExpressID, it uses the `ifcLoader.ifcManager.getItemProperties()` method to obtain the target component's name, type, GlobalID, label, and geometric dimensions. Finally, it loads and displays the target component's name, type, GlobalID, ExpressID, label, and geometric dimensions using the IFCLoader. An example of the obtained component attributes is shown below. Figure 4 As shown.

[0079] The component information acquisition module enables browsing and loading of target IFC model files. Based on the precise component picking technology of ray projection, the dynamic information panel displays key data such as the name, category, GlobalId, label, and geometric dimensions of the target component in real time, and links with the view highlight to form an efficient workflow of "click to understand".

[0080] In this embodiment, the performance of the component information acquisition module is optimized by setting the following settings:

[0081] The `geometry.toNonIndexed()` method is used to convert indexed geometry on the target component into non-indexed geometry. When dealing with mouse ray detection geometry, it is necessary to detect the ray along with each triangle. For indexed geometry, the actual vertices need to be found first using the index, which increases computational complexity. However, by directly using non-indexed geometry, the vertex data of each triangle is readily available, greatly simplifying the construction of the boundary volume hierarchy (BVH) and the ray detection process.

[0082] The `geometry.computeVertexNormals()` method is used to recalculate the normal vectors of the vertices of the triangle. When using the `toNonIndexed()` method mentioned above, the originally shared vertices are now copied. These new vertices will lose their original normal information, or the normal information will no longer be correct. Therefore, this method is used to fix the lighting display problems caused by geometry transformation.

[0083] The `geometry.boundsTree = new MeshBVH(geometry)` method is used to construct a boundary volume hierarchy for the geometry. An IFC model may contain hundreds of thousands or even millions of triangles. When performing ray detection, it is necessary to detect the intersection of the ray with each triangle, which is an O(n) linear operation with extremely poor performance. BVH, on the other hand, is a spatial partitioning data structure that organizes all triangles into a tree structure. When performing ray detection, it first checks whether the ray intersects with the bounding box of the root node. If it intersects, it continues to check the intersection with its child nodes. This process is repeated downwards until only a few triangles that may actually intersect with the ray are detected. This reduces the detection complexity from O(n) to O(log n), resulting in a performance improvement of hundreds or even thousands of times for large IFC models.

[0084] This invention aims to free users from reliance on specific software interfaces and technical jargon. Users can achieve precise component location, efficient attribute lookup, intelligent relationship analysis, and automated processing of complex tasks in IFC model files simply through intuitive and flexible natural language commands. Furthermore, the lightweight IFC-Viewer module developed based on this invention effectively addresses the need for local access to IFC model files and real-time metadata awareness, further enhancing the practicality and coverage of the technology. Through this disruptive "human voice as command, dialogue as control" model, this invention significantly improves the accessibility, parsing efficiency, and ease of operation of BIM model data, fundamentally lowering the application threshold and unlocking data value, providing a core driving engine for the intelligent upgrading of the construction industry.

[0085] like Figure 5 As shown, this embodiment also provides a working method for an IFC model cognitive agent based on the MCP protocol, including:

[0086] Step 1: Obtain the processing requirements of the IFC model file submitted by the user in natural language;

[0087] Step 2: Use a large language model to analyze the processing requirements based on context and confirm the processing method for the target IFC model file;

[0088] Step 3: Based on the analysis process, the filesystem-Server module is called to retrieve the target IFC model file from the file system, and then the IFC-MCP-Server module is called to perform query analysis, data modification, and / or automatic processing on the target IFC model file; and the processed target IFC model file will be stored back to the file system by the filesystem-Server module.

[0089] Step 4: Feed back the processed target IFC model file to the user in the required format; and during the overall operation of the IFC model cognitive agent, load and display the target IFC model file, GLTF model file, and GLB model file through the IFC-Viewer module so that the user can compare and review them.

[0090] The cognitive agent and its working method based on the MCP protocol IFC model provided in this embodiment have the following beneficial effects:

[0091] This IFC model cognitive agent, developed based on the IFC standard and MCP protocol using an MCP Server, deeply integrates LLM and breaks through the traditional BIM operation logic with a "dialogue-as-a-service" model. It enables rapid parsing and intelligent processing of components, attributes, and spatial relationships in IFC model files through natural language commands, increasing model interaction efficiency by more than 10 times and reducing professional-grade BIM operation time from hours to minutes. While reducing the learning cost of traditional software by 90%, it provides highly agile support for model review and emergency maintenance of large-scale infrastructure. In the design phase, architects and engineers can quickly complete scheme comparisons, specification reviews, and model compliance checks through direct dialogue, compressing professional operations that originally required hours to minutes. In the construction phase, site managers can directly query component information, locate installation positions, or simulate construction sequences using mobile terminals, greatly improving on-site decision-making efficiency and collaborative accuracy. In the operation and maintenance phase, facility managers can quickly retrieve equipment parameters, trace maintenance records, or simulate emergency scenarios through natural language commands without pre-installing professional software, providing agile support for the intelligent operation and maintenance of large-scale facilities. This technology substantially restructures the interaction paradigm of building digital assets, laying the technological foundation for the intelligent upgrading of the industry.

[0092] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0093] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A cognitive agent device based on the IFC model of the MCP protocol, characterized in that, include: Intelligent Hub, IFC-MCP-Server Module, IFC-Viewer Module; The intelligent hub includes a conversational AI framework built on a large language model, which is used to understand the user's processing requirements for IFC model files proposed in natural language by combining context, and to call the filesystem-Server module and IFC-MCP-Server module to create a workflow according to the user's requirements; and to add and apply the IFC-MCP-Server module and IFC-Viewer module to the intelligent hub. The filesystem-Server module includes an open-source filesystem service, which is used to retrieve the target IFC model file from the file system according to the calling command; The IFC-MCP-Server module is built on the FastMCP high-performance framework and the ifcopenshell underlying computing engine. It is used to query and analyze, modify data, and / or automatically process target IFC model files according to the calling instructions. The IFC-Viewer module is built on a modern web technology stack based on Vue3 + Three.js and is used to load and display target IFC model files, GLTF model files, and GLB model files. The IFC-Viewer module includes a local model loading module, a GLTF or GLB model file loading module, and a component attribute acquisition module; The local model loading module is used to obtain the target IFC model file in the local folder through the webkitdirectory property, and use the URL.createObjectURL() method to create a temporary access URL, which is then passed to the IFCLoader loader. The target IFC model file is then loaded and displayed based on the IFCLoader loader. A GLTF or GLB model file loading module, used to load and display GLTF or GLB model files using the GLTFLoader loader in Three.js; The component attribute acquisition module is used to obtain the vertex indices of the triangles on the target component detected by the mouse ray in the target IFC model file. Based on the vertex indices of the triangles, the `faceIndex` property of the triangle is called to obtain the indices of the triangles. The first vertex index of the triangle is obtained by multiplying the triangle index by 3. The `geometry.attributes.expressID.array[vertexIndex]` method is used to obtain the ExpressID of the target component. Based on the ExpressID, the `ifcLoader.ifcManager.getItemProperties()` method is used to obtain the name, type, GlobalID, label, and geometric dimensions of the target component. Finally, the name, type, GlobalID, ExpressID, label, and geometric dimensions of the target component are loaded and displayed using the IFCLoader loader.

2. The IFC model cognitive agent device based on the MCP protocol according to claim 1, characterized in that, The IFC-MCP-Server module includes a query analysis module, a data modification module, and an automatic processing module; The query and analysis module is used to query the geometric properties, engineering properties, containment system, and type hierarchy of target components in the target IFC model file; The data modification module is used to modify the cross-model operation scenarios of target components in the target IFC model file, batch update the attributes of target components in the target IFC model file, and delete components in the target IFC model file; The automatic processing module is used to perform integrated geometric and semantic transformation of the target IFC model file, as well as structured export.

3. The IFC model cognitive agent device based on the MCP protocol according to claim 2, characterized in that, In the query analysis module, querying geometric attributes includes querying the coordinates and spatial location of components; querying engineering attributes includes querying the name, information, material composition, fire resistance rating, and load-bearing characteristics of components; the query inclusion system includes querying the space where the component is located and all components within that space; the query type hierarchy system includes querying all components corresponding to a category, querying components made of concrete, and querying the type of each component.

4. The IFC model cognitive agent device based on the MCP protocol according to claim 2, characterized in that, In the data modification module, cross-model operation scenarios for modifying target components in the target IFC model file include copying components from the target IFC model file to other IFC model files; Batch updating the properties of target components in the target IFC model file includes batch standardizing the names of target components in the target IFC model file.

5. The IFC model cognitive agent device based on the MCP protocol according to claim 2, characterized in that, In the automatic processing module, the geometric and semantic integration transformation of the target IFC model file includes exporting the target IFC model file as a GLB model file; the structured export of the target IFC model file includes exporting the target components in the target IFC model file as a CSV model file.

6. The IFC model cognitive agent device based on the MCP protocol according to claim 2, characterized in that, The workflow of the IFC-MCP-Server module includes: Tool method design includes designing tool methods for query analysis, data modification, and automatic processing. Each tool method includes a method name, input parameters, return value data type, text comments of statements in the method, and method body; the input parameters are the calling instructions. Creating an MCP-Server instance includes constructing the MCP-Server instance by calling the FastMCP method in the mcp.server.fastmcp library; Registering utility methods includes adding @mcp.tool() before the utility method or registering the utility method to the MCP-Server instance using mcp.add(); Run the service, including running an MCP-Server instance and selecting the underlying transport method, which may be stdio or SSE protocol.

7. The IFC model cognitive agent device based on the MCP protocol according to claim 1, characterized in that, It also includes settings to optimize the performance of the component property retrieval module, including: The geometry.toNonIndexed() method is used to convert indexed geometry on the target component into non-indexed geometry to simplify the construction of the boundary volume hierarchy. The normal vectors of the vertices of the triangle face are recalculated using the geometry.computeVertexNormals() method to fix the lighting display problem caused by the geometry transformation; The geometry.boundsTree = new MeshBVH(geometry) method is used to construct a boundary volume hierarchy for the geometry in order to reduce the complexity of ray detection.

8. The IFC model cognitive agent device based on the MCP protocol according to claim 1, characterized in that, Conversational AI frameworks include Cherry-Studio, Claude, Fast-agent, and Dify.

9. A method for operating an IFC model cognitive agent device based on the MCP protocol as described in claim 1, characterized in that, include: Step 1: Obtain the processing requirements of the IFC model file submitted by the user in natural language; Step 2: Use a large language model to analyze the processing requirements based on context and confirm the processing method for the target IFC model file; Step 3: Based on the analysis process, call the filesystem-Server module to obtain the target IFC model file from the file system, and then call the IFC-MCP-Server module to perform query analysis, data modification, and / or automatic processing on the target IFC model file; Step 4: Feed back the processed target IFC model file to the user in the required format; and during the overall operation of the IFC model cognitive agent, load and display the target IFC model file, GLTF model file, and GLB model file through the IFC-Viewer module so that the user can compare and review them.

Images