Resin MCP server

Connect your client

Pick your client. The server is remote, so there is nothing to install: point the client at the server URL and sign in. The first connection opens a Resin consent screen where you choose what the agent may see and do.

What agents can do

Each area below comes with something you can say to your agent as it is written.

1. Design from a description

   Describe what you need and Resin starts a new project. It asks what it needs to know, chooses real parts, and draws the schematic.

   Ask your agent

   “Design a USB to CAN adapter in Resin: USB-C, an isolated CAN FD transceiver, and 3.3 V logic.”

2. Change an existing design

   Ask for a change in plain words. Resin decides which parts to use and how they connect, makes the change, and says what it changed.

   Ask your agent

   “In my usb-can project, add a CAN transceiver on CAN1 with decoupling and termination, and bring the bus out on J2.”

3. Read a design and write firmware from it

   The agent reads the circuit: every component, every net, and the net on each pin. From that it writes pin maps and compares firmware with the board.

   Ask your agent

   “Which MCU pin is SPI1_SCK on in the usb-can project? Generate a C header with all pin assignments.”

4. Find parts and create library parts

   The agent searches the parts catalog by keyword or part number and gets package, price, and stock. When Resin has no drawing for a part yet, the agent can have its symbol and footprint created in the project library. Creating a part uses AI credits.

   Ask your agent

   “Find a 3.3 V LDO in SOT-23 that is in stock, and create the library part for it if Resin does not have one.”

5. Run the checks and the design review

   The electrical checks and the layout checks are free and answer at once, in sentences. The design review reads the whole schematic and saves its findings to the review panel in Resin. It uses AI credits.

   Ask your agent

   “Run the checks on my sensor board and tell me what needs fixing. Then run a quick design review.”

6. Lay out a board

   Ask for the layout in plain words. Resin plans the board, places the parts, and routes it. The layout keeps working in the background and reports into the conversation. The agent follows the layout, passes on any question Resin asks, and can stop it. Resin can undo a finished layout when you ask.

   Ask your agent

   “Lay out the usb-can board. Keep J1 and J2 where they are, and tell me which nets are still unrouted when it is done.”

7. Import a design from another tool

   Resin gives the agent a short-lived upload address and the agent sends the file there. A design or a board from another tool becomes a new project. A library file is added to the library of a project you name.

   Ask your agent

   “Import hardware/usb_can.PcbDoc into Resin as a new project and give me the link.”

8. Export manufacturing files

   The agent asks Resin for the manufacturing package of a board and gets a download address that is valid for 15 minutes. Other exports, such as the bill of materials on its own, work the same way.

   Ask your agent

   “Export the manufacturing files for the usb-can board and save them in ./fab.”

How it works

Your agent sends a message to Resin with chat. It reaches the same Resin you talk to in the chat panel of the app. Resin makes the engineering decisions and answers in plain text.

• Conversations continue. The next message about a project goes to the conversation your agent already has there, so Resin remembers what was said. The agent can also start a new conversation.
• The conversation shows up in the chat panel in Resin, under your name. You can read it there and carry on yourself.
• A change applies to your project straight away. Open the project in Resin and the change is there.
• Every change is recorded as a version, and a reply from Resin says when a version was saved before the turn. To undo a change, open Version Control in Resin and restore an earlier version.
• Long tasks are waited on. A turn of a conversation, a design review, creating a part, a board layout, and the manufacturing package can take minutes. The agent calls wait until the answer is ready, and stop ends a task.
• Answers end with a link that opens what they are about in Resin: the sheet, the board, or the project.
• Things are named the way you name them: a project by its name, a component as U3, a net as +3V3, a pin as U3.7, a part by its manufacturer part number.
• The helper tools do what is asked and never guess. A helper that lacks something says exactly what to supply. Only chat makes engineering decisions.

A worked example

You ask your agent for a change. The agent passes the request to Resin with chat and gives you the reply.

Your agent calls chat

{
  "project": "USB CAN",
  "message": "Add a CAN transceiver on CAN1 with decoupling and termination, and bring the bus out on J2."
}

Resin answers

Resin: I added U4, a TCAN1042VDRQ1 CAN transceiver, on sheet CAN.
TXD and RXD go to the CAN1 pins of U1. C12 (100 nF) decouples the 5 V supply.
R7 (120 Ω) terminates the bus, and CANH and CANL go to J2 pins 2 and 3.
Questions:
  1. STB is tied to ground, so the transceiver is always on. Should a pin of U1 control it instead?
Changed: Added U4, C12 and R7 on sheet CAN. Connected CAN_TX, CAN_RX, CANH and CANL.
A version was saved before this turn, so it can be undone from Version Control.
Charged: 6.4 credits. This conversation so far: 6.4.
Open: https://resineda.com/w/…/projects/…?show=schematic&sheet=CAN

When Resin needs longer

Resin is working on it.
  Choosing parts
  Drawing the schematic
Call wait for the reply.

The reply says what Resin did, asks what it still needs to know, states what was charged, and ends with a link that opens the change in Resin. Nothing asks you to confirm a charge: an agent can spend only on a connection you gave the ai:spend permission, and never past the credit limit you set for it. get_usage shows what has been spent. When a turn takes longer than the call waits, the agent gets the progress so far and calls wait for the reply.

Authentication and permissions

Sign-in is OAuth 2.1. You approve each connection on a Resin consent screen, where you choose the workspace, which projects the agent may see (all of them or the ones you select), and which permissions it gets.

• An agent can never do more than you can.
• A connection limited to chosen projects cannot create a new project.
• Workspace admins can turn agent access off and turn personal access tokens off.
• Every tool call is recorded. You can see recent activity and disconnect any app in Settings > Agents & API.

Credits

Most of what an agent does is free. The tools that use AI credits ask first and say what they charged.

• Tools that use AI credits: chat, generate_part, and run_ai_review. Every other tool is free. Calls are rate-limited.
• Talking to Resin, generating a library part and running an AI design review use AI credits. Each is charged for the AI its run actually uses. That is not known until the work is done, so there is no price list and no estimate.
• Your agent is not asked to confirm a charge. A connection may spend only when you gave it ai:spend, and never past the credit limit you set for it.
• You are charged once, when the work succeeds, and the answer states the exact credits charged. Work that fails or is stopped costs nothing.
• get_usage reports what the connection has been charged, what is left of its limit, what the workspace has used and has left this month, and what each recent paid task cost.
• A board layout that Resin starts from a chat message is not charged per message: it uses a share of the workspace's monthly AI allowance, which you can change by asking in chat, and it counts toward the workspace total, not toward the connection's limit.
• An answer from Resin also states what the conversation has cost so far.

Tools

The server offers these tools, grouped by what they are for. Every tool takes a few plain arguments and answers in plain text. Each one is described the way the server describes it to your agent.

Plugin and skills

A Resin plugin for Claude Code adds two skills: a base skill that teaches an agent how to work with Resin through these tools, and a firmware skill for pin maps. The plugin is coming to the plugin directory.

Security and limits

• Sign-in uses OAuth 2.1 with PKCE.
• Tokens are scoped to this server.
• Calls are rate-limited.
• Upload and download addresses are signed and stop working after 15 minutes. Resin checks your permissions again when one is used.
• Text that comes from datasheets, part listings, or imported files is marked as untrusted, and agents are told not to follow instructions inside it.
• Supported protocol versions: 2025-03-26 through 2026-07-28.

Troubleshooting

Find the symptom and apply the fix.