AI Assistants

This is an experimental beta feature. AI assistants can make mistakes. Review what they change, and create a revision before a large edit so you can roll back.

Connect an AI assistant to your harness.design account and it can work on your harnesses with you: look one up, explain how it’s wired, add connectors and wires, assign parts, and answer questions about how the software works. Anything it changes appears in your editor live, while you watch.

The connection uses MCP (Model Context Protocol), the standard way an AI assistant connects to an outside service. harness.design runs an MCP server, so any assistant that speaks MCP can connect to it.

Set it up from Main Menu → Preferences → AI.

The assistant signs in as you, so it sees exactly the harnesses you see: your own, plus any shared with you or belonging to your team. It can never reach anyone else’s work.

Connecting an assistant

Open Preferences → AI and expand the assistant you want to connect. Each one shows whether it’s currently Connected, along with the steps to set it up.

Wherever an assistant asks for the MCP server URL, use:


https://app.harness.design/mcp

undefined

In the AI preferences, click Add connector to Claude. Claude opens with the connector details already filled in. Confirm it, then sign in with your harness.design account when prompted.

To add it by hand instead, open Customize → Connectors in Claude, add a custom connector, and paste in the MCP server URL above.

Claude is available in the browser and as a desktop app, and both work the same way once the connector is added.

undefined

In ChatGPT, open Settings → Security and login and turn on Developer mode.
Open Settings → Plugins and click +. Name the connector harness.design, paste in the MCP server URL above, and sign in with your harness.design account when prompted.

The image ChatGPT asks for is optional, so you can skip it.

Starting a session on a harness

Once an assistant is connected, an Edit with … entry appears for it in the Main Menu. Choosing it opens that assistant with the harness you’re currently working on already loaded, ready for your first question.

For assistants with a desktop app (currently Claude), an Open in Browser / Open in Desktop app switch in the AI preferences controls which one that entry opens.

You don’t have to start from the menu. You can also just ask the assistant directly (“list my harnesses”, or “open my Engine Bay harness”) and it will find it.

Example prompts

You talk to an assistant in plain English. Here are some example prompts to give a sense of what you can ask for. Use them as starting points and swap in your own parts and harnesses.

Adding parts to your library

Add the full catalog of Deutsch DTM connectors to my parts library. Be sure to include required accessories such as wedgelocks, and include the compatible contacts.

Adding a device

Add the Link G5 Voodoo Neo 4 ECU to my harness. Use the correct connectors from TE, referring to Link’s documentation.

Importing from Excel

Import this harness into harness.design. It was created in Excel.

Attach the spreadsheet to your message and the assistant reads it and recreates the harness.

Support questions

In harness.design I have a connector with different gauge wires coming into it. How do I assign different contact sizes to a single connector?

Assistants answer questions like this by reading these documentation pages.

Everything an assistant writes goes through the same validation and the same plan limits as the editor. On the Free plan the 100-connection limit per harness still applies, and creating revisions still requires Pro.

Working with an assistant

A few things that make a real difference to the results:

Name components the way the editor does. Component IDs like C1, W3 and S2 are exactly what the assistant uses to find things, so “change W12 to blue” or “what’s connected to C4 pin 7?” lands precisely.
Say which harness, or say “the harness I have open”. The assistant can see which one you’re working on and will confirm the name back to you.
Ask for a checkpoint before a big change: “create a revision first, then rewire the ground circuit.” It won’t create one unless you ask.
Be specific about parts. If you name a real part number, the assistant will look it up and create a part matching the actual specification. If you don’t, it follows whatever convention the rest of the harness already uses.
Watch it work. Because changes appear in the editor live, you can see each one land. If you don’t like a change, Undo works exactly as it does for your own edits. Each thing the assistant does is one step in your history. (Restoring a revision is the exception, just as it is when you restore one yourself.)

While an assistant is editing

When a connected assistant starts changing a harness you have open, an indicator appears in the editor bar (for example Claude is editing this harness…) so you always know when a change isn’t yours. It clears itself shortly after the assistant finishes.

Disconnecting

In Preferences → AI, expand the assistant and click Disconnect. It immediately loses all access to your account and harnesses until you connect it again.