The AEC sector’s longstanding drawback is the dearth of information circulation amongst processes and mission members. We’ve tried to repair it with shared file repositories, shared BIM fashions, and APIs that allow software program information alternate. Regardless of new AI instruments, the problem stays. However there could also be an answer.
Anthropic launched MCP (Mannequin Context Protocol) in November 2024 to allow AI programs to entry numerous information sources in a standardized means. It has gained speedy acceptance amongst builders, and a few AEC tech builders are already exploring it.
Might MCP be the lacking hyperlink to make AEC information accessible and helpful for everybody? In any case, it has been known as the “USB-C of AI apps.”
The idea
Presently, AI instruments like ChatGPT primarily function primarily based on the knowledge supplied within the immediate and their very own coaching information. If you need them to entry reside mission information, BIM fashions, or inner firm instruments, you want customized integrations or proprietary APIs.
Conventional APIs are made for builders, not AI fashions. They require extra layers of interpretation earlier than AI can make the most of them successfully. MCP supplies a regular “language” (message format) for AI fashions to request information or carry out actions, and for information suppliers to reply.
With MCP, you possibly can:
- Expose chosen information (e.g., from a BIM database, value estimating software, or mission schedule)
- Give the AI restricted actions (e.g., run a question, validate a mannequin, generate a report)
- Management what it might probably and may’t do
- Do all this with out retraining the AI
MCP servers
To make use of MCP, an AI agent connects to an MCP server. It’s software program that acts as a bridge between the AI and a number of of the info supplier’s programs (API, mission administration system, file storage, and many others.) and is aware of the way to talk with them.
MCP servers can present three foremost varieties of capabilities: sources (information for purchasers to learn), instruments (capabilities that the agent can name), and prompts (templates for workflows).
The MCP server affords machine-readable descriptions of its capabilities, together with names, parameters, and explanations. This helps the AI agent perceive what capabilities can be found and the way to use them.
The info supplier chooses what capabilities the server exposes, for instance, “getProjectStatus” or “fetchLatestIFCFile.” The AI can’t transcend that.
MCP purchasers
Desktop AI assistants like ChatGPT and Claude act as hosts for MCP purchasers; they run built-in purchasers that hook up with exterior or native MCP servers. The host spins a consumer occasion for every server.
While you ask ChatGPT (the host) for project-specific information, it evaluates the out there MCP servers and selects the proper software.
The MCP consumer then sends the server a structured request, the server executes the logic (for instance, querying a BIM database), and returns structured outcomes. ChatGPT interprets these outcomes and weaves them into its natural-language response.
Potential AEC purposes
You’ll be able to envision many purposes the place MCP would turn out to be helpful. Listed here are just some examples:
- Building course of monitoring: “Which elements marked ‘fabricated’ within the BIM are nonetheless not put in on-site?”
- Data reuse over tasks: “What subcontractors have delivered related MEP programs on time and on finances within the final 5 hospital tasks?”
- Regulatory compliance: “Which supplies specified within the BIM mannequin have Environmental Product Declarations (EPDs)?”
- Product information retrieval: “Discover different air dealing with items that meet the necessities of the designed unit.”
- Cross-disciplinary coordination: “Summarize all modifications to the structural mannequin since final week.”
- KPI monitoring: “How a lot embodied carbon is within the present design, and the way does that evaluate to the unique goal?”
Making information actionable
To reply such questions, the agent should accumulate information from a number of programs. This solely works if the info is standardized throughout sources, utilizing constant identifiers, items, and schemas.
The outcomes may be mixed both by the AI consumer or by the MCP server. Consumer-side becoming a member of supplies flexibility, however server-side composition is mostly sooner, extra constant, and simpler to handle.
I beforehand mentioned data graphs. A graph supplies semantic that means throughout programs, giving a related, standardized view of the mission. A graph, for instance, connects the BIM factor, the ERP part, and the purposeful objective, and MCP supplies the bridge between the AI agent and the graph.
Software program assist is rising
Since MCP was solely not too long ago launched, few AEC instruments presently assist it natively. Nonetheless, there are a number of MCP server implementations for Autodesk Revit, developed by group contributors and AI-focused corporations. The MCP servers talk with Revit utilizing the Revit API via WebSocket or HTTP connections.
Fragments is an open-source BIM format and toolkit designed to effectively retailer, show, navigate, and edit giant quantities of BIM information on any gadget. An early adopter, Helen Kwok, has developed an MCP server to transform IFC recordsdata into Fragments format, load Fragments, and question BIM information by class.
From an idea to enterprise use
Though MCP stays in an exploratory section inside AEC, it’s positioned to turn out to be a vital expertise for integrating AI into workflows. As an alternative of people prompting an LLM, AI brokers can invoke MCP instruments autonomously, which permits automation of workflows, for instance, for provide chain administration.
By monetizing your structured mission information via AI-ready MCP servers, you may make income per question whereas creating a powerful incentive to put money into larger information high quality.
MCP continues to be evolving and requires enhancements in safety, governance, and sure technical capabilities. Nonetheless, I like to recommend staying present and experimenting with MCP-enabled workflows at present in safe environments.
