What is MCP:
MCP, or Model Context Protocol, is a standard protocol for establishing communication between different tools and the client. Previously, for communication between tools and their use, a separate communication protocol had to be developed to enable applications to use various tools. Anthropic introduced MCP as a standard protocol, making communication between tools and clients (such as Claude Desktop) much simpler.
For better understanding of this concept, consider a simple example. Suppose that, in order to use a computer, we need a mouse. If each company producing a mouse created its own special connector for connecting to the computer, then using mice across different computers would become a serious problem, and we could not use different mice with a single computer. However, with the design of USB connectors, and with all companies adopting that standard, there is no longer a need for various connectors, and it is possible to use any mouse with any computer. In this analogy, the mouse is the tool, the USB connector represents MCP, and the computer represents the client or consumer.
Nowadays, with the daily progress of technology—especially artificial intelligence and LLMs—and the expansion of their use in different fields, the AEC industry has not been excluded. Specialists in this field have begun working to create a connection between artificial intelligence and the industry. The world of BIM is one such field that has benefited from the advances of artificial intelligence, and every day new tools and plugins are developed to increase speed and efficiency. In this post, we will become familiar with MCP in Revit software.
For using different LLMs in Revit software, an interface is required to enable communication between Revit and the LLM. Through this interface, tools are made available to the LLM, allowing it to access Revit element data and execute different codes within Revit. One of these communication methods is MCP.
For this reason, various tools have been developed that, by using the MCP protocol, provide a collection of tools to the LLM so that it can perform different tasks in Revit.
One of the clients that supports the MCP protocol is Claude Desktop. By supporting MCP, Claude Desktop can establish a bridge between its Chabot and Revit. In parallel, different plugins have been created that act as the required tools for the LLM. By connecting to MCP, they enable the transfer of Revit data to the LLM. Developers can also produce dedicated MCP servers, packaging specific tools as needed and making them available to the LLM.
By installing Claude Desktop software and connecting an MCP Server, there is no need to export information from Revit and send it to a Chabot, or to receive code from the Chabot and then manually run and debug it in Revit. In this method, because an external software is required for executing commands and obtaining outputs from Revit—here, Claude Desktop—you do not have a UI in Revit for performing tasks. Instead, everything is sent to Revit by Claude Desktop through the MCP protocol.
All of these processes are automatically generated using the tools available in the MCP Server. In this way, the LLM is provided with a collection of tools and instructed to use them whenever needed, while the remaining steps are planned and executed by the LLM itself.
What is BIMLOGIQ Copilot?
BIMLOGIQ Copilot is a specialized LLM for code generation, optimized specifically for Autodesk Revit. It is context-aware, enabling it to intelligently interpret and respond to user prompts within the Revit environment. As mentioned earlier, many tools have been created for using LLMs in the AEC industry, particularly in Revit software. One of these tools is a Chabot called BIMLOGIQ Copilot. BIMLOGIQ Copilot directly addresses the limitations of general-purpose LLMs in the AEC industry, particularly in Revit software. Unlike typical chatbots that simply translate user requests into code, BIMLOGIQ Copilot is tuned to understand the Revit environment and workflows. It reduces the common errors caused by incomplete prompts, lack of Revit-specific training, or misinterpretation of element names and parameters. By integrating domain knowledge, structured instructions, and curated examples, it ensures that user requests are executed reliably, covering both common and complex tasks that general LLMs fail to perform consistently.
Unlike MCP, which only provides a set of tools for the LLM to select from, BIMLOGIQ Copilot employs more advanced methods. Instead of presenting a limited set of tools to LLM, it first analyzes the user’s prompt with different instructions, evaluates it within the Revit domain, and uses multiple samples and prior examples to accurately interpret the request. Based on the specific prompt, it supplies the LLM with the necessary information—such as element types and required parameters—and includes instructions for accurate code generation to minimize errors.
With this method, the LLM is provided with knowledge of the Revit API and with instructions for proper implementation of the task. At the same time, the data and content required for correct execution are obtained from the Revit project. Finally, the code generated by the LLM is reviewed through advanced methods and, if it passes a series of criteria, it is executed in Revit.
Unlike the previous method (MCP), in this method you have a plugin inside Revit itself, and there is no need to use external software. As a result, you have a UI within Revit for executing commands and communicating with artificial intelligence.
Comparison of the performance of the two tools:
Because of the difference in the implementation of these two tools, there are naturally differences in their performance. The first method, using MCP in Revit, relies on the tools provided by the developer. However, in many cases the user requires tasks for which no tool has been designed in advance. In such cases, it becomes necessary for the LLM to generate code, send it to Revit, and receive the result (assuming a tool for executing code in Revit is available).
In these cases, due to the presence of many unknown parameters and the use of an LLM not specifically trained for Revit code, numerous errors occur. Sometimes, even after extensive back and forth between Revit and the client, these errors—often simple ones—remain unresolved, and the task fails.
On the other hand, the BIMLOGIQ Copilot plugin uses an LLM tuned specifically for executing Revit tasks. It also employs various tools to inject Revit knowledge into the LLM, prevent known errors, and correct different errors before executing the code in Revit. As a result, it provides more optimized performance.
For example, when using the prompt “Split selected ducts into 1000mm fixed lengths,” MCP cannot handle it directly. The user must provide an additional hint to use MechanicalUtils.BreakCurve() for splitting MEP ducts. Furthermore, in some cases MCP does not convert numbers into Revit’s internal units, requiring explicit instructions to perform the conversion.
In general, simple and medium-level tasks can be executed with MCP—particularly tasks that do not involve modifying the Revit model and only require data extraction. However, tasks of higher complexity, or those requiring special techniques, generally fail with MCP.
In contrast, tasks of simple, medium, and moderately complex difficulty can be executed with BIMLOGIQ Copilot, except in cases limited by the Revit API or when the task requires vision, complex operations, or human judgment.
Nevertheless, each of these tools has its own strengths and weaknesses, which are discussed below.
Feature | MCP (Model Context Protocol) | BIMLOGIQ Copilot |
|---|---|---|
Execution Environment | Requires external software (Claude Desktop) | Built-in Revit plugin |
Interaction with LLM | Provides tools directly to the LLM, which decides how to use them | Analyzes the user’s prompt, filters relevant data, and guides the LLM for accurate code generation |
Setup Complexity | Requires installing dependencies, server setup, and configuration | Simple installation as a Revit add-in |
Task Suitability | Works best for simple tasks and data extraction | Handles simple, medium, and moderately complex tasks |
Error Handling | High error rate due to LLM not being trained on Revit code | Reduces errors with Revit-specific knowledge and pre-validation |
Execution Speed | Very fast for simple tasks | Slightly slower for very simple tasks (around 30 seconds) |
Report Generation | Supports HTML/CSS for advanced, visual reports | Limited to text or Excel reports |
Web Search Capability | Yes (via Claude Desktop) | No (limited to Revit environment) |
Reusability of Commands | Not supported | Supported — commands can be saved, reused, and pinned to the ribbon |
Command Sharing | Not available | Available — share commands with colleagues |
Built-in Command Library | None | Yes — includes public ready-to-use Revit commands |
@Mention Elements in Chat | Not supported | Supported — directly reference project elements during chat |
C# Coding Support | Limited and error-prone | Full built-in environment to write, compile, and run C# code |
Complex Task Handling | Weak, often fails without pre-built tools | Stronger — succeeds in most cases except API limitations or tasks requiring human judgment |
Strengths of each product:
Naturally, each tool has its own strengths and weaknesses, and the selection of the appropriate tool must be based on specific needs. Below, the strengths and weaknesses of each tool are examined.
MCP in Revit:
1- The ability to produce attractive and eye-catching reports by using HTML and CSS code: In the MCP method, because external tools are used—here, Claude Desktop—it is possible to take advantage of their capabilities. Since Claude Desktop includes an HTML page renderer, it is possible to generate customized and visually appealing reports from the Revit model according to specific needs.
2- Support for web search: Another advantage of MCP in Revit is the search capability of Claude Desktop. Therefore, if specific analyses are required based on external resources, this capability allows information to be extracted from Revit, relevant standards or resources to be searched on the internet, and, finally, the Revit output data to be analyzed or the desired task to be executed in Revit based on those resources.
3- High execution speed: If the desired task is performed by one of the tools available to the client, it is executed at high speed, typically within a few seconds. Tasks with a very low level of difficulty are also executed quickly.
4- Open protocol for switching between AI models: This open protocol enables interaction with Revit using different AIs. In other words, it is possible to switch between Claude, GPT, or other models without requiring changes to the structure.
BIMLOGIQ Copilot plugin:
1- No need for specialized knowledge of coding: One of the strengths of this plugin is that it does not require specialized programming knowledge. By simply chatting, you can perform the desired task. In case an error occurs, by explaining the error in English to BIMLOGIQ Copilot, you give it the opportunity to debug the code again.
2- Ability to save commands for reuse: Although performing tasks may take slightly more time compared to MCP, after successful execution you can save the task as a command. By doing so, a tool is created with the appropriate inputs, and the command is stored. You can also add it as a tool to the ribbon bar, so that in future projects you can run it again without re-entering the prompt, simply by filling in the command inputs.
3- Ability to edit command: Another ability of this plugin is the option to edit saved commands. If a saved command needs to be modified to improve part of its performance or to add functionality, it can easily be edited by chatting and then saved again as a new command.
4- Sharing commands: Another strength of this plugin is the ability to share commands with others. If you execute a successful command and save it, you can share it with colleagues or friends so they can add it to their own list of commands.
5- Ability to mention elements during chat: Another strength of this plugin is the ability to reference elements directly in chat. If you need to use a specific element, such as a type or a family, you can simply refer to it by writing @ in the chat. This reduces the error rate in task execution and eliminates the need to manually search the project for element names.
6- Having a library of public commands: BIMLOGIQ Copilot provides users with a list of public commands. These commands, produced by the company for all users, include a collection of useful tools in Revit.
7- Ability to search within project documents: Another key strength of BIMLOGIQ Copilot is its ability to search within project documents while executing a task. This ensures that if the names or types specified in the chat contain spelling mistakes or are incomplete, BIMLOGIQ Copilot attempts to identify and match them with the correct entries before passing them to the LLM.
Example: If a user prompt is:
“set ofset from level of all air terminals in space with name of Workshop to 2800mm”
Even though the parameter name contains a typo (e.g., “ofset” instead of Offset from Host), BIMLOGIQ Copilot automatically searches the project, identifies the correct parameter name (Offset from Host), and applies it.
This functionality is particularly valuable because users often overlook the importance of using the exact parameter, family, type, or other project names. Instead, they may write approximate or incorrect names. The plugin resolves this by intelligently mapping user input to the correct project terms.
8- Environment for executing C# codes: For users specialized in programming, an environment is available for executing C# code directly. This allows users to write, compile, and execute C# code without the need for additional tools.
Conclusion
MCP and BIMLOGIQ Copilot each offer distinct approaches for integrating LLMs with Revit. MCP provides flexibility through its open protocol, external tool integration, and features such as report generation and web search. However, its reliance on generic LLMs and external interfaces often results in limitations when executing complex or Revit-specific tasks.
BIMLOGIQ Copilot, by contrast, is designed specifically for Revit. It addresses the shortcomings of general-purpose LLMs by embedding domain knowledge, offering an in-application plugin with a user interface, and providing advanced capabilities such as reusable commands, element referencing, intelligent error handling, and a dedicated C# execution environment. As a result, it delivers higher reliability and efficiency for both simple and complex Revit tasks.
Ultimately, the choice between MCP and BIMLOGIQ Copilot depends on the user’s needs: MCP offers openness and interoperability across AI models, while BIMLOGIQ Copilot delivers a focused, optimized, and practical solution for Revit users seeking accuracy and productivity within the BIM workflow.
