devbord speaks MCP: let your AI assistant work the board

devbord speaks MCP: let your AI assistant work the boardPoint Claude or any other MCP client at devbord.com/mcp and it can find, create and update cards, comment, tick off checklists and triage error issues. 22 tools, one URL, the API keys you already have.2026-10-10Your AI assistant already writes the code. Then it politely asks you to go and move the card to Review yourself, as if you were the junior. That ends today: devbord has an MCP server.One URL, the keys you already haveMCP (Model Context Protocol) is the open standard that lets an AI assistant use tools outside its chat window. devbord now is one of those tools, at a single address:```
https://devbord.com/mcp
```There is nothing to install and no second set of credentials. The server takes the same API keys as the REST API.Connect in one commandClaude Code:```bash
claude mcp add --transport http devbord https://devbord.com/mcp \
--header "Authorization: Bearer $DEVBORD_KEY"
```Any other client that reads a `.mcp.json`:```json
{
"mcpServers": {
"devbord": {
"type": "http",
"url": "https://devbord.com/mcp",
"headers": { "Authorization": "Bearer db_xxxxxxxx_…" }
}
}
}
```Create a key per project under Project settings › Integrations › API keys, or for the whole workspace under your workspace › API. Right after you create one, the connect command is shown with the key already filled in. Copy, paste, done.What the assistant gets22 tools, in four groups:- **Read**: list boards, get a board with its lists, labels, members and custom fields, list and search cards, read recent activity, and open a card in full: description, checklists, custom fields, attachments, latest comments, branches and pull requests.
- **Write**: create, update, move and archive cards, post comments (mentions notify people), add checklists and tick off items, attach links, set custom fields, create lists.
- **Git**: link a branch or pull request to a card, unlink it, and refresh its CI, reviews and commits from GitHub.
- **Errors**: list the issues from error tracking, read the stack trace of one, turn an issue into a card, and resolve, ignore or assign it.Things you can now say out loud- "What is due this week on the Website board?"
- "Go through the unresolved errors, make cards for the three that hurt most and put them at the top of Backlog."
- "I just opened pull request 214. Link it to the login bug card and move that card to Review."
- "Tick off 'write tests' on card 42 and comment what changed."The boring parts we did on purposeHanding a language model the keys to your board deserves some paranoia. Ours:- **The key decides.** A board key sees one board, a workspace key sees all of them. Give the assistant a read-only key and the tools that write are not even on its list.
- **Nothing deletes.** There is no delete tool. Removing a card means archiving it, and you can restore that.
- **Retries do not make duplicates.** `create_card` with `issue` returns the open card of that error issue instead of a second one, and with `idempotency_key` the same call gets the same answer for 24 hours. An assistant that retries after a timeout leaves one card, not five.
- **No back door.** Every tool is a call to the public API with your key, so the same permissions, plan and limits apply. Tool calls count towards the key's 300 requests a minute.Fine print- The MCP server is part of the Premium plan, like the rest of the API.
- Transport is Streamable HTTP: POST only, no session. There is no OAuth sign-in, so the client has to be able to send its own `Authorization` header.The full reference is in the [API documentation](/api-docs), under AI assistants (MCP).devbord · Plans · News · API documentation · What we can and cannot seedevbord requires JavaScript to work.