Design perfboards with an AI agent

The agent reads the saved board: every part, every net, and the pins each net connects. From that it writes pin definitions for your firmware instead of working from a half-remembered schematic or a photo of the bench.

• Pin maps from the real layout. Ask for a pins.h, a config struct or a device tree fragment, and each entry traces back to a net on the board.
• Kept in sync. Move an LED to another pin in the editor, then ask the agent to read the board again and update the firmware. The board revision tells it exactly which version it wrote against.
• Checked both ways. Hand it existing firmware and it can list every pin the code uses that the board does not connect, or every net on the board the code never touches.
• Layout context, too. Structured inspection gives the agent part positions and board geometry, so it knows which connector sits on which edge and which parts share a region of the board.

Starting from nothing works too. Describe the project and the agent creates a board, picks parts from the component library or draws a custom footprint from the part's datasheet, places them, defines the nets, previews routes and commits the wires and solder bridges.

The interesting part is that pin assignment becomes a free variable. When the agent owns both sides, it doesn't have to route a three-bend wire across the board to reach the pin the firmware happened to pick. It can use a pin next to the part instead and change one line of code. That is a trade a person rarely bothers to make by hand, and it adds up on a hand-soldered board where every bend is a bend you make with pliers.

• Placement first. Connected parts go close together with their leads lined up, and the agent can ask for a checked placement suggestion before it routes anything.
• Wiring you can build. Route previews prefer a short solder bridge, then a straight wire, then one bend, then two. Wires land on the free hole beside a pin, the way you would solder them.
• Constraints stay put. Pins with a fixed job (an analog input, a hardware I2C or SPI bus, a boot strap) stay where the datasheet puts them. Everything else can move to make the layout simpler.
• Safe to experiment. The agent can save a checkpoint before trying a different arrangement and roll back if it gets worse.

You stay in the loop the whole time. Open the board in the editor, move things around, and the agent picks up your changes the next time it reads the board.

• Read my board "Plant monitor" and write a pins.h that matches it.
• I moved the status LED on the board. Read it again and update the firmware.
• Check this sketch against my board and list every pin that does not match.
• Design a small perfboard for a dev board, three buttons and an I2C display, then write the firmware for it. Pick pins that keep the wiring short.
• Save a checkpoint, then try moving the connector to the other edge and tell me whether the wiring gets simpler.

PerfBoard is a remote MCP server with OAuth sign-in, so there is nothing to install. Add it to your client with this URL, sign in with your PerfBoard account, and choose which workspaces the agent can reach.

https://perfboard.app/mcp

Step-by-step instructions for Claude, ChatGPT, Claude Code, Codex and Cursor or VS Code are on the AI Agents page in the app. Agent access needs PerfBoard Pro and the MCP subscription; designing by hand stays free.

Can an AI agent design a perfboard layout?

Yes. Connect an MCP client such as Claude, ChatGPT, Claude Code, Codex or Cursor to PerfBoard and the agent can create a board, place parts from the component library, draw custom footprints from a datasheet, define nets, preview routes and commit wires and solder bridges. It runs through the same board engine as the PerfBoard editor, so the result is an ordinary board you can open, check and change by hand.

Can an AI agent read my perfboard layout to write firmware?

Yes. The agent reads the saved board: every part, every net and the pins each net connects. It can turn that into pin definitions for your firmware, keep them in sync when you move something in the editor, and check existing firmware against the way the board is actually wired.

Which AI tools work with PerfBoard?

Any client that supports remote MCP servers with OAuth sign-in. The AI Agents page in the app has step-by-step connection instructions for Claude, ChatGPT, Claude Code, Codex and Cursor or VS Code, plus the server URL for anything else: https://perfboard.app/mcp.

Can the agent change boards I did not ask it to touch?

Only inside what you approve. When you connect, you choose the permissions and which workspaces the agent can reach, and you can revoke a connection at any time from the app. Every edit is checked against the board revision it was based on, and the agent can save a checkpoint before a risky change and roll back to it.

Do I still need to check the datasheet?

Yes. An agent can copy pin numbers and labels from a datasheet into a footprint, and it is told to verify them, but a design rule check cannot prove a pinout is right. Check the pinout of every IC and discrete part against its datasheet before you wire or power the board.

What does agent access cost?

Designing in PerfBoard is free. Connecting an AI agent over MCP needs PerfBoard Pro and the MCP subscription, both managed from the AI Agents page in the app.