Claude × Revit MCP Implementation | How to Control BIM with Natural Language

For architectural designers and BIM managers who want to "have Claude talk to Revit and generate walls" or "extract quantities from a BIM model using natural language," this guide lays out the current state of Revit MCP and the fastest path to getting it running. We cover both options — the OSS revit-mcp and the built-in MCP server in Revit 2027 — along with how to connect from Claude Desktop and where it fits into real-world workflows.

Summarypowered by Claude
Article Summary via AI Chatpowered by Claude
結論powered by Claude

Revit MCP is a system that bidirectionally connects Revit and Claude via the Model Context Protocol, enabling natural language instructions for reading, editing, and code compliance checking of BIM models. Two systems coexist: the OSS revit-mcp/revit-mcp (MIT) and the built-in MCP server that ships standard with Revit 2027. The former works with versions prior to 2026 but requires plugin installation, while the latter requires no installation and connects directly from Claude Desktop.

The basic setup route is Node.js 18 or later + revit-mcp-plugin + editing claude_desktop_config.json in Claude Desktop. Once you add the path to the built JS file under mcpServers, a hammer icon appears on the Claude side and the connection is complete. Since it has write access to the model — including element retrieval, parameter updates, and C# code submission — it is safest to test on a backed-up .rvt file rather than a production file.

The greatest real-world impact comes from investigation-type, one-off tasks such as "code compliance checks," "solar panel placement calculations," and "floor inventory extraction across all levels." A division of labor is emerging where high-frequency, repetitive operations remain the domain of pyRevit, while complex exploration and conditional aggregation go to MCP. Both Autodesk officially and the community recommend that all output be reviewed by a qualified architect.

目次 (8)

What Is Revit MCP — The Common Protocol Linking BIM Models and Claude

Revit MCP is the collective name for servers that implement Anthropic's Model Context Protocol (MCP) within Revit. As background: BIM (Building Information Modeling) refers to a design methodology that integrates geometry, structural members, and building systems into a single 3D model. MCP is a common standard for connecting AI to external applications — much like USB-C, the goal is "plug it in and it just works."

Once an MCP server is running in Revit, natural language instructions from Claude Desktop — such as "Tell me the total length of walls on the second floor" or "Tag only fire-rated 60-minute doors in red" — are translated directly into Revit API commands and applied to the model. The fundamental difference from conventional AI integration is that it directly accesses the live .rvt file without going through screenshots or IFC exports (Autodesk Platform Services Blog).

Two Implementation Options — OSS revit-mcp and Revit 2027 Built-In

As of May 2026, there are two main routes for controlling Revit from Claude.

Type Provider Supported Revit Requirements License
revit-mcp (OSS) revit-mcp / mcp-servers-for-revit Revit 2019–2026 Node.js 18+ + plugin installation MIT
Built-in MCP server Autodesk Revit 2027+ Claude Desktop only Commercial license

The OSS version is available on GitHub at revit-mcp/revit-mcp, where revit-mcp-plugin runs persistently on the Revit side and communicates with the MCP server over a socket. Revit 2027 ships this functionality built in — users simply install the Claude Desktop app (free) and launch Revit to connect (Bimsmith Blog).

What You Can Do — Read, Create, Edit, Analyze

The core capabilities of the revit-mcp family fall into five categories:

  • Model reading: View information, elements, family types, parameter values, geometry, project information
  • Element operations: Create point/line/face-based elements, place grids and levels, delete elements, update parameters
  • Quantity takeoffs: Material quantity calculations, statistics by element category, CSV/Excel export
  • Visualization: Color-coding by condition, tagging, filter application
  • C# code submission: Send arbitrary Revit API code as a string and execute it on the spot

The last capability — C# execution — is powerful, but since write permissions pass through directly to the Revit API level, a strict backup-first policy is essential.

Setup Steps — revit-mcp + Claude Desktop

The fastest route to running the OSS version on Revit 2024/2025/2026 is these five steps:

  1. Install Node.js 18 or later (verify with node -v)
  2. git clone https://github.com/revit-mcp/revit-mcpnpm installnpm run build
  3. Place revit-mcp-plugin on the Revit side (place the .addin file in %AppData%\Autodesk\Revit\Addins\<version>\)
  4. Add the following to Claude Desktop's claude_desktop_config.json
{
  "mcpServers": {
    "revit-mcp": {
      "command": "node",
      "args": ["C:\\path\\to\\revit-mcp\\build\\index.js"]
    }
  }
}
  1. Restart Claude Desktop — if revit-mcp appears in the hammer icon at the bottom right of the input field, the connection is complete

For Revit 2027's built-in version, steps 1–3 are unnecessary. Simply open Claude Desktop while Revit is running and it will appear as a connection option.

Real-World Commands with Revit 2027 Built-In MCP

Practical examples reported by Autodesk and Bimsmith include the following:

  • "List the floor inventory for all levels" → Automatically generates a report in a Word document
  • "Check code compliance against the building code for the building's location" → Lists non-compliant elements
  • "Place solar panels on the roof at maximum capacity" → Regenerates in seconds
  • "Modify a Python script to match the parameters of the selected wall" → Completed in combination with pyRevit

As even Autodesk officially states, "AI is wrong sometimes." The key to improving accuracy is a workflow where you point out errors within the chat and run re-analysis, treating the first response as a starting point rather than a final answer.

Caveats — Accuracy, Scope, and Final Review by a Qualified Architect

There are three constraints to understand before deploying Revit MCP on real projects:

  • Accuracy is not perfect: Code compliance, egress routing, and structural judgments all require review by a qualified architect
  • Context limits have real-world effects: For large projects, prompts scoped to "second floor only" or "exterior walls only" are more efficient
  • Write operations are irreversible: Back up journal (Revit's operation log file) and .rvt (Revit project file) first, then test on a separate file

The older OSS repository (mcp-servers-for-revit/revit-mcp) was archived in February 2026, and the project has since migrated to the revit-mcp/revit-mcp monorepo. Links in tutorials and older articles may be broken, so always get the latest version from the main repository.

Division of Labor — When to Use pyRevit vs. MCP

pyRevit (a Python-based automation add-on for Revit) and Revit MCP are complementary, not competing:

  • Where pyRevit excels: High-frequency tasks that repeat the same operations daily, batch processes standardized within an organization, routine work you want turned into UI buttons
  • Where Revit MCP excels: Investigation tasks where conditions vary by project, one-off exploration, aggregation where you want to specify conditions flexibly in natural language

Archilabs.ai's report concludes that "the strongest teams don't abandon pyRevit — they use both." The emerging best practice is to use MCP for feasibility studies in early design phases and pyRevit for repetitive production work during drawing submission phases.

Conclusion — Revit 2027 Makes "Talking to BIM" the Standard

Revit MCP is the first serious implementation that allows reading and writing BIM models in natural language, and with its inclusion as a built-in feature in Revit 2027, "conversing with BIM software" has become an industry-standard capability. Whether to get a head start with the OSS version on pre-2027 Revit or wait for the 2027 upgrade depends on your deployment scale, but the dramatic productivity gains for investigation-type, one-off tasks have already been proven.

If you already have a Claude Pro plan and the Claude Desktop app installed (free), you can get started without any additional AI usage fees. We recommend preparing one practice .rvt file and starting with "What is the total floor area of this building?".

参考になったら ♡
Clauder Navi 編集部
@clauder_navi

Anthropic の Claude / Claude Code を中心に、日本のエンジニア向けに最新動向と実務 を毎日発信。 運営方針 は メディアについて をご覧ください。