NavisCoord

English · Español

Turn thousands of Autodesk Navisworks clashes into explainable, verified
decisions your coordination team can schedule.

Install in five minutes ·
Latest release ·
What is verified ·
Security model ·
Ask a question

NavisCoord is an open-source, local MCP server plus a native Navisworks add-in.
It is built for BIM/VDC coordination, not generic UI automation. Model data
stays on the workstation.

Not affiliated with Autodesk. A licensed installation of Autodesk
Navisworks Manage is required for the live add-in.

The problem it solves

A federation can produce 5,000–50,000 clashes. Those are not 50,000 separate
decisions. NavisCoord filters expected contacts, collapses duplicate geometry,
reconciles levels, ranks actionable issues, detects systemic root causes and
builds meeting-sized work packages.

Install — beginner path

Requirements: Windows, Python 3.10+, and Navisworks Manage 2024, 2025 or 2026.
You do not need Visual Studio or the .NET SDK when using a release.

1. Install the Navisworks add-in

Close Navisworks. Download
Install-NavisCoord.ps1, then run:

powershell -ExecutionPolicy Bypass -File .\Install-NavisCoord.ps1

The installer detects installed Navisworks versions, downloads the correct
assets from the latest release, verifies SHA256SUMS.txt, validates the ZIP
and publishes only NavisCoord-owned files with rollback.

Prefer doing it manually? Download the ZIP for your version from the
latest release
and extract it to:

%APPDATA%\Autodesk\Navisworks Manage <version>\Plugins\NavisCoord\

2. Add the MCP plugin

Claude Code:

/plugin marketplace add HorizunGroup/naviscoord-mcp
/plugin install naviscoord-mcp@horizun-navis

For Codex, add this repository as a plugin marketplace and enable
naviscoord-mcp@horizun-navis.

The launcher creates a private, versioned Python environment on first start.
It does not modify your normal interpreter and installs only hash-locked
dependencies.

3. Paste the first safe prompt

Verify NavisCoord without modifying or saving the model. Then discover and
analyze the open Navisworks document, identify root causes, and produce a
coordination work plan in read-only mode.

Full screenshots-free walkthrough and troubleshooting:
Quick start.

What it can do

• Discover the real model/property structure instead of assuming one BIM
  standard or language.
• Build selection/search sets and clash matrices from a versioned profile.
• Run tests, group results and prepare viewpoints through guarded workflows.
• Filter non-actionable contacts and collapse repeated clashes.
• Reconcile inconsistent storey names from geometry.
• Rank severity and explain why an issue matters and which side should move.
• Detect systemic elevations, missing penetrations, repeated typologies and
  congested zones.
• Produce work plans, narratives, PDFs, CSV star schemas and Revit/BI handoff
  files.
• Target one of several simultaneously open Navisworks sessions explicitly.
• Run long operations as jobs with honest progress and cancellation semantics.

The analysis engine also works offline against an exported JSON file, without
Navisworks:

python -m naviscoord analyze export.json --profile my-project.json

First workflows

Read-only analysis:

navis_health
→ navis_discover
→ navis_analyze
→ navis_root_causes
→ navis_work_plan
→ navis_pdf_report

Operational coordination:

navis_audit_models
→ navis_configure
→ navis_run
→ navis_group_levels
→ navis_save_as

Ask for navis_capabilities before relying on a mutation. If multiple
Navisworks instances are open, choose one with navis_target; NavisCoord
refuses to guess.

Why it is different

Coordination intelligence, not remote control

The deterministic server does the large computation and gives the language
model compact evidence. If an LLM has to read clashes one by one, the design
has already failed.

Explainable by construction

Weights, tolerances, discipline rules, criticality and movability live in a
project profile. The score is arithmetic rather than learned, so a coordinator
can challenge it, change it and reproduce it.

Honest writes

Every mutation uses a dry-run/commit contract, explicit session and document
identity, idempotency where retries matter, and post-commit re-reads whenever
Navisworks exposes a reliable getter. Unsupported verification is reported as
unsupported — never upgraded to success because a call returned.

Local security boundary

The C# bridge listens on loopback, uses a per-session bearer token stored under
an explicit Windows DACL and marshals bounded work onto Navisworks' UI thread.
Output paths are confined; profile checksums and document fingerprints bind
state to the intended model.

See the complete security model and report
vulnerabilities through GitHub private advisories, not issues.

Compatibility

One source is compiled once per Navisworks release because Autodesk ships a
version-specific API assembly. Autodesk binaries are never redistributed.

Architecture

MCP client (Claude / Codex / compatible host)
        │ stdio
        ▼
Python server ── analysis · planning · reporting · output policy
        │ authenticated loopback HTTP
        ▼
Native C# add-in ── UI-thread dispatch · Navisworks API · document re-reads
        │
        ▼
Active Navisworks Manage document

The split is deliberate: analysis stays deterministic and testable without a
license; only API-bound work enters Navisworks.

Evidence, not test-count marketing

Public CI exercises Python 3.10–3.14, both supported MCP major versions,
packaging, schemas, security contracts and the Navisworks-free C# logic.
Release builds are deterministic, binary artifacts are scanned for private
absolute paths, dependencies are hash-locked, and mutations are tested against
their common envelope.

Navisworks itself does not run on a public hosted runner. Live checks are
recorded by version and model type in Testing, including
what was not exercised. A green hosted-CI badge is not presented as proof
that a live model was opened.

Run the repository-safe suites:

python -m pytest -q
dotnet run --project addin\NavisCoord.Tests -c Release
python scripts\build_artifacts.py --check
.\Install-NavisCoord.ps1 -SelfTest

Install from source

Close Navisworks, then:

git clone https://github.com/HorizunGroup/naviscoord-mcp.git
cd naviscoord-mcp
.\install.ps1
cd server
python -m pip install -e .

install.ps1 detects compatible local Navisworks installations and compiles
against each API. Building requires the .NET SDK 8+; the project supplies the
.NET Framework 4.8 reference assemblies.

Manual MCP configuration starts from .mcp.json.example.

Documentation

Community

• Ask usage and installation questions in
  Discussions.
• Search existing issues before reporting a reproducible bug.
• Read CONTRIBUTING.md before a pull request.
• Use private security reporting for tokens, paths or
  vulnerabilities.
• If NavisCoord saves your team a coordination session, star the repository
  and share the before/after counts — sanitized, never with client data.

The project benchmark is intentionally public. Current gaps such as a real
sanitized demo, signed beginner installer, PyPI distribution and official MCP
Registry publication are tracked in docs/BENCHMARK.md, not
hidden behind the badge row.

License

MIT — see LICENSE. Autodesk trademark and third-party notices are in
NOTICE. No Autodesk assembly, model or client dataset is included.