The Model Context Protocol (MCP) is an open standard that lets an AI assistant connect to tools and data sources through a common interface, so the same assistant can look up records, run searches and take actions without a custom integration for each system. For construction and infrastructure teams, MCP is most useful once project content from every common data environment (CDE) sits together in one governed place, such as your own Azure tenancy, where an MCP server can expose it safely.
Where MCP came from
MCP was published as an open specification in late 2024 to solve a practical problem. Every AI application needed its own connector to every data source, which meant a growing tangle of one-off integrations. MCP separates the two sides. A data source or tool is wrapped once as an MCP server, and any AI application that speaks MCP can use it as a client. Many AI assistants, developer tools and agent frameworks now support it, and organisations are writing their own MCP servers for internal systems.
The protocol describes three main things a server can offer. Tools are actions the AI can call, such as "search documents" or "list open RFIs". Resources are pieces of content the AI can read, such as a file or a register. Prompts are reusable instructions that shape how the AI approaches a task. The AI application decides when to call them, and the server decides what it will allow.
Why CDE data is a hard case for MCP
On paper, you could point an MCP server at each CDE's API and let an AI assistant query them all. In practice that runs into problems quickly. Each platform has its own API, permission model, rate limits and metadata. Some platforms restrict the API calls available to third parties. Content held in one vendor's platform may be owned by the client, the contractor or a joint venture, with different rules for each. And an assistant firing live queries at five vendor platforms during a busy week is a load and security concern that most IT teams would rather avoid.
There is also a quality problem. AI answers are only as good as the content they draw on. If drawings in Autodesk Forma use one revision scheme and the same drawings in Oracle Aconex use another, an assistant has no reliable way to tell which is current. The data needs to be gathered, normalised and indexed before an AI can reason across it with confidence.
The pattern that works
The more dependable pattern puts a copy of the project record in storage your organisation controls, then exposes that copy to AI. It runs in three layers.
Content from each CDE is replicated into your Azure tenancy, in the region you choose, with metadata mapped to a common schema.
That content is indexed for search, including OCR of scanned drawings and PDFs, so it can be retrieved by meaning as well as by keyword.
AI applications reach the index through a governed interface. That might be a built-in assistant, your own application, or an MCP server your team writes over the index.
With this arrangement, the MCP server talks to one well-structured store under your policies instead of five vendor APIs. Permissions, logging and retention are handled once. The AI gets consistent metadata, and the vendor platforms are left to do their day jobs.
What an MCP server over CDE data might offer
If your team decides to build one, the tools it exposes should match the questions people actually ask. Typical examples include searching documents by meaning across every project, retrieving the current revision of a drawing by number, listing transmittals sent to a party within a date range, summarising correspondence on a topic with citations, and returning document control statistics for a project. Each tool should respect the caller's permissions and return references back to the source file so people can check the answer.
Keep the first version small. A handful of well-designed read-only tools will deliver more value, and attract fewer security questions, than a broad set that can write back to live platforms.
Security questions to settle first
MCP is a protocol and does not decide what is safe. Before connecting any assistant to project records, agree who can use it, which projects and document classes it can see, whether it can take actions or only read, where prompts and responses are logged, and which AI models are approved. For organisations with Security of Critical Infrastructure (SOCI) obligations, keeping the data, the index and the model inside your own tenancy is usually the simplest way to answer those questions. Our article on running AI across multiple CDEs in your own Azure covers the architecture in more detail.
MCP and the Utopiaverse™
The Utopiaverse™ builds the governed copy that makes MCP practical. CDE Sync™, CDE Backup™, CDE Migrate™ and CDE Archive™ bring content from Forma, Aconex, ProjectWise, Procore, SharePoint and other platforms into your own Azure, so every project and every CDE across the organisation sits in one connected record. AI Insights then answers questions across that record with RAG and OCR search, citing the source of every answer, while the file viewer and dashboards give people direct visibility of documents and document control status across the portfolio.
Because the record lives in your tenancy, your team is free to add its own AI on top, including an MCP server built on the same data, without asking each CDE vendor for access. The result is one place to see and question everything your organisation holds, whichever platform it was created in.
See the Utopiaverse™ in the monthly live demo, or talk to the team about bringing your CDEs together for AI.