Connect Claude or another assistant
You can connect your own AI assistant to your models. It can read them and propose changes, and every change asks you first. Claude, ChatGPT, Grok, Cursor, or any assistant that speaks the Model Context Protocol (MCP) can connect.
You pay your AI provider. The app pays for no AI and sees nothing of your AI account.
This page covers the hosted app at https://app.mutinybim.com. If you run your own copy, use your own address in its place. For running the tool server on your own computer, the full tool list and the scripting kit, see Connect an AI assistant.
The address
The address your assistant connects to is the app's address with /mcp on the end:
https://app.mutinybim.com/mcpThe Connect an assistant page (in the top bar, and in the bar above the editor) shows this address with steps for each assistant.
Add it to your assistant
Add MutinyBIM as a custom connector (an MCP server) and paste the address.
Claude (claude.ai or Claude Desktop): Settings → Connectors → Add custom connector, paste the address, then Connect.
ChatGPT: Settings → Apps & Connectors → Create (developer mode), paste the address, choose OAuth, then Create.
Grok: add a custom connector (MCP server) and paste the address.
Claude Code:
bashclaude mcp add --transport http mutinybim https://app.mutinybim.com/mcpCursor: add this to
~/.cursor/mcp.json:json{ "mcpServers": { "mutinybim": { "url": "https://app.mutinybim.com/mcp" } } }
These menu names belong to each assistant and may change. If a step doesn't match what you see, look for "custom connector" or "MCP server" in your assistant's settings.
Sign in and press Allow
When you connect, your assistant opens a MutinyBIM sign-in page. Log in, and the app asks, with your assistant's name: Let "Claude" use your models? There is no key to copy.
The page says what the assistant can do:
- It can read your models and the ones shared with you.
- It can propose changes. Every change asks you first, and models shared with you to view stay view-only.
- It can't see your password, your settings or your AI keys, and it can't share or delete models.
- You can sign it out any time under Connect an assistant.
The name at the top is the one the assistant gave itself. Only allow it if you asked your assistant to connect to MutinyBIM a moment ago. Press Allow, or Don't allow.
After you allow it, the assistant is listed under Assistants signed in on the Connect an assistant page, with when it signed in and when it was last used. Sign out there cuts its access to your models at once.
An assistant acts for you and nobody else: it reaches your models and the ones shared with you, nothing more. Its changes, and the editor's list of who is here, carry the name it registered with, marked as an agent.
How approvals work
Nothing an assistant does lands in a model until you approve it.
- In the assistant. If your assistant has its own approval box, it asks you there. You see what will change and approve or decline. The assistant can't answer for you.
- On the Approvals page. If your assistant can't ask you itself, the change waits on Approvals in the app. The Approvals link in the top bar shows how many are waiting. Open one to see exactly what changes, then press Approve or Decline.
- In the editor and on every page. While something waits, a notice shows it with Approve, Decline and See exactly what changes.
The assistant waits about a minute for your answer. After that the request stays on Approvals for 30 minutes and lands the moment you approve it. After 30 minutes it is dropped.
A script your assistant runs only previews at first. It changes the model only after you approve, and it lands as one undo step. Scripts run on the app's server in a sealed box with no network.
Try it: build a small office
The project ships four example scripts. The small office builds two floors of a 40 x 30 ft office: exterior and interior walls, doors and windows, rooms found from the walls (open office, conference room, office), a 9 ft ceiling in each room, and a plan of each floor on its own sheet.
On My models, make a new blank model. Call it "Small office".
In your connected assistant, ask:
In my MutinyBIM model "Small office", run the small-office example script. The model already has Level 1: use it rather than making a new one.
A blank model starts with a level named Level 1, and the example makes its own. Run as written, it stops with "There is already a level named Level 1". The example's own notes say how to use a level the model already has, and the assistant can follow them.
The assistant fetches the example, checks it and runs a preview. Nothing changes yet.
It asks you to approve: in its own approval box, or on Approvals. Look at what will change, then approve.
Open the model. You should see the walls, rooms and sheets on two levels.
You can also paste the script's text into the assistant and say "run this script". The other examples build a rooftop unit with its ducts, a plumbing riser and an electrical room; see the examples list.
Tool families
When an assistant connects to /mcp, it starts with the core tools, ten at most. Their main job is writing, checking, previewing and landing scripts, plus a few reads to look at the model first. Most work is done by scripts.
Other tools come in families. An assistant turns one on when a job needs it; you can ask it to ("turn on the electrical tools"). Or connect a family directly at its own address, https://app.mutinybim.com/packs/<family>/mcp, for example https://app.mutinybim.com/packs/electrical/mcp.
The families are model, drafting, ducts-pipes, electrical, plumbing, views-sheets, coordination, groups-phases, parts, calcs and schedules. What each one covers, and every tool in it, is in the tool server reference, generated from the tool server itself.
Every change from any family asks you first, the same as the core.
Run the tool server on your own computer
Some assistants only start local tools. For those, the Connect an assistant page has Or run the tool server on your own computer: make an agent key with Make a key, and run the command it shows. The key is shown once; anyone with it can edit your models, so treat it like a password. Turn off a key to cut its access. See Connect an AI assistant for the details.
Ask AI or connect an assistant?
- Ask AI, in the editor, answers questions from a summary of the model with your own key. It does not change the model. See Getting started.
- A connected assistant reads the model and proposes changes you approve.