NavisCoord MCP Server
io.github.HorizunGroup/naviscoord-mcp
Connect Autodesk Navisworks to Claude and ChatGPT for AI-driven clash analysis, root-cause investigation, and coordination planning.
What is the NavisCoord MCP server?
NavisCoord is an MCP server that integrates Autodesk Navisworks with AI clients like Claude Desktop and ChatGPT to automate clash detection, triage, and coordination workflows. It runs as a native Windows add-in paired with an MCP server, enabling teams to analyze model clashes, identify root causes, generate work plans, and produce coordination reports without manual review of thousands of clash rows.
NavisCoord bridges Navisworks and AI by letting you ask natural-language questions about model clashes and coordination issues. Instead of manually reviewing clash reports, you can ask the AI to rank issues by priority, identify root causes, propose solutions, and prepare coordination meeting materials. The server runs locally on your Windows workstation with no API keys or external services required.
How to install NavisCoord
Copy-paste configuration for popular MCP clients.
Tools & capabilities
Tools this server exposes to the agent.
Clash analysis and ranking— Analyze open Navisworks models, identify active clashes, and rank coordination issues by deterministic priority scores with explanations.Root-cause investigation— Group spatially bounded clashes, collapse duplicates, and generate hypotheses about underlying causes with measured evidence.Change detection— Compare model snapshots, track persistent issue identities, and report what changed since the last delivery.Work-plan generation— Create work packages, assign trade ownership, and generate narrative coordination reports with PDF images.Handoff file creation— Export structured data with authoring element IDs for downstream tools like Revit and Power BI.Targeted model writes— Apply reviewed coordination decisions to the Navisworks model with document/revision checks and read-back verification.Company profile configuration— Load coordination standards to control disciplines, category mappings, tolerances, priorities, and coordination rules.
Use cases
- Automatically rank and prioritize coordination issues in large federated models to focus meeting time on critical conflicts.
- Identify root causes of clashes and propose targeted changes that resolve multiple occurrences at once.
- Generate coordination meeting agendas and work packages with trade assignments and narrative explanations.
- Compare model versions to track which issues are resolved, new, or persistent across deliveries.
- Export clash data and element IDs to Revit or Power BI for downstream coordination workflows.
NavisCoord MCP server FAQ
NavisCoord is an MCP server that connects Autodesk Navisworks to AI clients (Claude Desktop, ChatGPT Desktop, Codex, Claude Code) to automate clash analysis, root-cause investigation, and coordination planning. It runs locally on Windows with no external APIs or subscriptions required.
Yes, NavisCoord is open-source under the MIT license. You need a licensed copy of Autodesk Navisworks Manage (2024, 2025, or 2026) and a compatible AI client (Claude Desktop, ChatGPT Desktop, Codex, or Claude Code).
Download the `.mcpb` file from the GitHub releases page and follow the client-specific instructions in the documentation. The package includes the standalone server and all Python dependencies; no separate Python installation is needed.
Run the desktop-plugin installer, then open Plugins → Personal → NavisCoord → Install. The plugin runs locally on your workstation without connecting to OpenAI Platform or requiring a tunnel.
No. NavisCoord runs entirely on your Windows workstation. It does not upload the complete model or send telemetry to Horizun services. Requested outputs (model names, element properties, coordinates, images) are sent only to your chosen AI client.
Autodesk Navisworks Manage 2024, 2025, and 2026 are supported. The standalone analysis engine can also process saved exports without a live Navisworks installation.
README (reference)
Source of truth, from the repository.
NavisCoord
Your next coordination meeting starts with a decision, not 10,000 clash rows.
Connect Autodesk Navisworks to ChatGPT Desktop Work, Claude Desktop, Claude Code and Codex. Ask what matters, inspect the evidence, and build a coordination plan.
Get started · Release evidence · Report a problem
1.0.0 is public: stable download · official MCP Registry entry · Smithery.
“Analyze the open model. Show the most important coordination issues, explain their priority, and prepare the work plan. Show me the proposed changes before applying them.”
NavisCoord pairs a native Navisworks add-in with an open-source MCP server. The engine runs on your Windows workstation. The desktop runtime includes its Python dependencies: no Python installation, API key or build tools are needed for the packaged local server. Your chosen AI client and Autodesk licence remain separate products.
What you can do
| Ask for | Get back |
|---|---|
| “What needs attention first?” | Ranked issues with deterministic scores and the reasons behind them |
| “Are these really separate problems?” | Spatially bounded groups, duplicate collapse and traceable original clashes |
| “Could one change address several occurrences?” | Root-cause hypotheses, measured evidence and every affected occurrence |
| “What changed since the last delivery?” | Persistent issue identities, saved snapshots and a comparison of changes |
| “Prepare the coordination meeting.” | Work packages, trade ownership, narrative reports and PDF images |
| “Connect this to Revit and Power BI.” | Authoring element IDs, structured handoff files and tables for downstream tools |
| “Apply the reviewed plan.” | Targeted writes with document/revision checks and read-back verification |
53 tools. One server. Four clients. Explore the tool guide.
Choose your client
| Client | Install |
|---|---|
| ChatGPT Desktop — Work | Run the desktop-plugin installer, then open Plugins → Personal → NavisCoord → Install |
| Codex | Use the same personal plugin, or register the executable with codex mcp add |
| Claude Desktop | Install the release .mcpb, which includes the standalone server |
| Claude Code | Install naviscoord-mcp@horizun-navis, or register the executable with claude mcp add |
All four connect to the local workstation. ChatGPT Desktop uses a local plugin; OpenAI Platform, a tunnel and a runtime API key are not part of this installation. See OpenAI’s desktop-plugin documentation.
Install the two pieces
- Navisworks add-in: close Navisworks and run
Install-NavisCoord.ps1. It detects your installed version, verifies the release checksums and backs up managed files. - Your AI client: install the desktop package or
.mcpbfrom Releases. Follow the client-specific instructions. - Open Navisworks Manage, load your federation and ask: “Check NavisCoord, identify the active document and analyze its clashes.”
Windows and a licensed Autodesk Navisworks Manage 2024, 2025 or 2026 are required for live coordination. The standalone analysis engine can also process saved exports without Navisworks.
Built to make the evidence inspectable
- Conservative filtering: a category name alone does not prove that a door, sleeve or slab overlap is intentional. Hosting/contact rules require relationship evidence.
- Freshness checks: changes to geometry, properties, saved selections or clash results invalidate the analysis used for derived writes.
- Complete execution: a folded decision retains its individual occurrences in write-back and Revit handoffs.
- Measured proposals: movement alternatives use element bounds. They are proposals to check against surrounding geometry and design constraints.
- Honest delivery comparisons: an issue missing from a later analysis is not automatically labelled “resolved.” Coverage, exclusions and source changes remain visible.
- Reproducible installation: pinned dependencies, artifact checksums, a standalone runtime and rollback-aware installation.
The release verification record distinguishes automated checks, Navisworks acceptance tests and client verification. A directory listing or a passing test suite is not a promise that software can never fail. External marketplace review is recorded separately from our tests.
Works with your coordination standard
Load a company profile to control disciplines, category/file mappings, tolerances, priorities and coordination rules. Start with the example profile. A model with missing discipline or level data is identified explicitly so the team can map it before relying on the ranking.
NavisCoord produces evidence and proposed coordination actions. Engineering approval and edits in the authoring application remain with the project team. Handoff files carry the IDs needed to continue with tools such as Horizun Revit MCP and Horizun PBI MCP.
Privacy Policy
NavisCoord does not upload the complete model or send telemetry to a Horizun service. Requested tool outputs can include model names, element properties, coordinates and images; your AI client receives those outputs. Read the privacy policy and security model.
Build and contribute
python -m pytest server/tests -q
dotnet run --project addin/NavisCoord.Tests -c Release
See Testing, Releasing and Contributing. Please report reproducible cases with sensitive project information removed.
If NavisCoord helps your team, share a reproducible coordination example and star the repository so other BIM coordinators can find it.
MIT · Built by HorizunGroup · Independent of Autodesk, OpenAI and Anthropic.
Related MCP servers
Microsoft Project
Audit, reschedule and author Microsoft Project schedules. No Java, no Project install.
Horizun PBI MCP
Build, audit and repair Power BI Desktop models and reports with DAX, TMDL and PBIR using plain language.
Horizun Revit MCP
AI-powered BIM automation for Revit 2023–2027: model, coordinate and deliver with 123 tools.
24 tools for evidence-graded clinical queries, hormones, peptides, longevity & drug interactions.
HostTracker
Website uptime monitoring: run checks from 300+ locations, manage monitors, alerts and incidents
Microsoft 365 & Azure admin: Graph, ARM, and the PowerShell cmdlets Graph does not expose.