DevvyBoard

Team workspace

AI-native project management

MCP project management — a board your AI agents can actually use

Your AI coding agent can write the fix, run the tests, and open the PR — and then it stops, because your project board has no door for it to walk through. The Model Context Protocol (MCP) is that door: an open standard that lets tools like Claude Code, Cursor, and Codex CLI call real product APIs as if they were built in. DevvyBoard ships a first-class MCP server, so the board your team works is the board your agents work.

One npx command, no clone, no build

Create a scoped API key on the Connectivity page, paste one block of JSON into your tool’s MCP config, and ask your AI to “list my DevvyBoard projects”. The developer docs have copy-paste configs for Cursor, Claude Code, and Codex CLI.

{
  "command": "npx",
  "args": ["-y", "devvyboard-mcp"],
  "env": { "DEVVYBOARD_API_KEY": "dbk_live_YOUR_KEY" }
}

19 tools, four surfaces

The board

List projects, read the full board (columns, categories, cards, priorities, deadlines), then create, update, move, and delete cards. “Move everything I finished today to Done” is a one-liner.

Ideas & roadmap

Triage the ideas backlog, add new ideas, promote the best ones onto the board, and publish or unpublish the public roadmap.

Chat

Post into team chat — a long-running agent can report “deploy finished, tests green” where your team actually reads, or summon @devvy.

The Blueprint

Read your architecture diagram as a clean text outline — or draw on it: add nodes and connections, link them to kanban cards. A snapshot is taken before every agent edit, so any AI change is one click to undo.

Built so you can trust an agent with it

  • Scoped, revocable keys — a read-only key can never change anything; per-key project access; one click to revoke from the dashboard.
  • Snapshots before every agent edit — Blueprint writes automatically save a restore point first. An AI change is always one click from undone.
  • “Extend, never reflow” — when an agent draws on your Blueprint, auto-layout places new content in free space and never moves a node a human placed.
  • Rate limits per organization — Free includes 300 requests and 50 writes per day; Pro lifts the limits.

Who else does MCP project management?

Honestly: a few, partially. Linear has a full MCP server for issues. ClickUp’s is in beta with daily call caps, Monday.com’s exists but is buried in API docs, and Jira has announced one that hasn’t shipped. Trello has none. DevvyBoard’s difference is surface area: agents reach the board and the chat and the ideas backlog and the roadmap and a Blueprint canvas they can literally draw architecture on.

Proof it’s real: the product map in our own workspace — 22 nodes, 28 edges, linked to shipped cards — was drawn entirely by an AI agent over this MCP server. Zero coordinates supplied.

Give your agent a board this afternoon

MCP works on the free tier — no paid plan required. Five minutes from API key to “list my DevvyBoard projects”.

MCP project management — a board your AI agents can use | DevvyBoard