Skip to Content

Start here

Connect your AI client

Open the live session to your own AI client over the Model Context Protocol. One click inside the application — no config file to edit.

What actually happens

Clicking Connector starts a small HTTP server on 127.0.0.1 — your machine only. A bundled bridge relays it to your AI client over standard input/output, the transport MCP clients expect.

In Revit the server is off until you start it; the other applications start it when they open, and Settings can turn that off. It stops with you or the host, and nothing keeps running afterwards.

The one-click setup

  1. Open Settings on the AEC Pilot ribbon tab, then the Setup Guide tab.

  2. Pick your client: Claude Desktop, Claude Code, ChatGPT / Codex, Cursor, Google Antigravity.

  3. Click the setup button. It writes its own entry into that client’s MCP configuration and nothing else.

  4. Restart the AI client so it re-reads its configuration, then ask it something about the open document.

The Setup Guide tab in Settings, listing the AI assistants found on this PC with a Configure button for each.

Settings → Setup Guide: the assistants found on the machine and what they’re set up for. Leave the multi-instance switch alone unless you drive several projects at once.

It will not trample your other MCP servers

It backs up the config file, adds only its own entry, and leaves your other servers as they were.

Which server name to expect

Each product has its own name, so all of them can be connected at once.

ProductServer nameDefault port
Revitaec-pilot7823
AutoCAD & Civil 3Daec-cad7900
Navisworksaec-navis7950
Inventoraec-inventor8000
3ds Maxaec-max8050
Robot Structural Analysisaec-robot8100

The port is configurable in Settings. Change it only if something else uses it — then re-run the setup so the client follows.

Checking it worked

Two places tell you, and they answer different questions.

In the host application, the ribbon shows Connected or Disconnected; Settings → About shows the address and how many tools it registered.

The About tab showing Connected, the localhost address and port, and the number of tools registered.

Settings → About. Connected, the port it is listening on, and how many tools it registered — if the count is there, the client has something to talk to.

In your AI client, the tools should be listed. Server but no tools? Start the connector from the ribbon. No server at all? Restart the client. More in Troubleshooting.

Several windows open at once

With several windows of one application open, tool calls go to the one you opened first — and every reply names it.

To target each window separately, turn on the multi-instance pool in the Setup Guide. Each window gets its own connector name.

Controlling what it can do

Settings → Tools lists every tool with a checkbox. Disabled tools are hidden completely — the assistant can’t call them or even see them.

The Settings window on the Tools tab, listing every tool with an enable checkbox and a read-only column.

Settings → Tools, here in Navisworks. The Read-Only column tells you which tools can change the document — a quick way to leave inspection on and switch editing off.

Use the Read-Only column: keep the readers, switch off the writers, and the assistant can answer anything and change nothing.

The code tool is the one to decide on: it compiles and runs C# against the live document, for anything the named tools don’t cover. Code is screened against a blocklist (files, processes, network, registry), and the feature can be switched off centrally. In the chat panel, Require approval before custom code previews it first.