Natural-Language Clash Queries
Ask in plain English: “show active duct-vs-structure clashes above 20 mm.” The server translates intent into Navisworks tool calls.
MCP server · Python · MIT · Windows / macOS / Linux · Autodesk APS compatible
No more filtering rulesets by hand or wrangling spreadsheet exports.
Navisworks MCP Toolkit turns any MCP-compatible AI client
into a co-pilot for clash detection, model review, and quantity queries —
straight out of your .nwf files.
Purpose-built for BIM coordinators, VDC engineers, and AI systems.
An MCP-compatible Python server that runs locally on Windows and exposes five Navisworks tools — clash listing, clash execution, clash export, model review, and model-item queries — to any MCP-compatible AI client via stdio transport. Built for BIM coordinators, VDC engineers, and automation pipelines powered by Autodesk Platform Services.
Every capability is a first-class MCP tool with a typed schema.
Ask in plain English: “show active duct-vs-structure clashes above 20 mm.” The server translates intent into Navisworks tool calls.
Stdio transport, stateless tools. Plugs into Claude Code, Claude Desktop, Cursor, and any MCP-conformant agent.
Wraps Autodesk Clash Detective, Model Review, and model-item queries. Supports .nwf and .nwd sources.
Single pip install. Runs locally over stdio — no cloud dependency, no data leaves your machine unless you want it to.
Three adapters (mock, XML, COM) covered by a deterministic test suite so you can trust returns without opening Navisworks.
Free 100 calls/month for evaluation. Usage-based pricing from €0.01 per tool call thereafter. Cancel whenever.
Install, register, query. Stdio transport — works anywhere MCP does.
01# 1 — install02pip install navisworks-mcp03 04# 2 — register in Claude Code / Desktop (.mcp.json)05{06 "mcpServers": {07 "navisworks": {08 "command": "navisworks-mcp",09 "env": { "NW_MODE": "navisworks", "NW_FILE_PATH": "C:/models/site.nwf" }10 }11 }12}13 14# 3 — ask in natural language15> show active clashes on level 3 between ducts and beams16 17⟶ clash_detective_run({ testName: "HVAC vs STRUCT", level: 3 })18⟶ { clashes: [ { id: "CL-001", x: 1240, y: 308, z: 45, ... } ] } Three steps. Each one leaves an audit trail in your MCP client's session log.
Install the Python MCP server. Point it at a .nwf or .nwd file.
Register the server in your MCP client's config. No cloud roundtrip.
Ask clash, review, or model questions in natural language. Get structured JSON back.
Free tier to evaluate. Metered thereafter. No seats, no tiers-shaped-like-handcuffs.