Concord Mcp

Concord MCP

[](https://www.npmjs.com/package/@concord-ai/concord-mcp) [](https://www.npmjs.com/package/@concord-ai/concord-mcp) [](https://github.com/Get-Concord-AI/concord-mcp/actions/workflows/ci.yml) [](https://www.npmjs.com/package/@concord-ai/concord-mcp) [](./LICENSE)

Shared work-state for coding agents. Concord MCP gives Claude Code, Codex, Cursor, and other MCP-capable coding assistants a shared work log. Agents can claim work, leave handoffs, and generate review-ready packets before opening PRs.

> ⚠️ Early and under active development. The v0 surface is three MCP tools: > claim_work, handoff, and review_ready.

Why

Agents work well in isolation, but their plans, assumptions, and decisions only become visible at PR time. Teams hack around this today with AGENTS.md, handoff files, worktrees, and custom scripts. Concord packages the smallest useful version of that: a local place for agents to record what they are doing while they do it.

Install

npm install -g @concord-ai/concord-mcp
concord install

concord install writes Concord's tool instructions into your client configs (CLAUDE.md, AGENTS.md, .codex/, .cursor/rules/). Then register the MCP server with your client and let your agent use the tools through MCP. Per-client setup:

• [Claude Code](./docs/claude-code.md)
• [Codex](./docs/codex.md)
• [Cursor](./docs/cursor.md)

> There is no universal /concord slash command — commands are client-specific. > Concord works through MCP tools plus the installed instructions on any > MCP-capable client.

The three tools

| Tool | When | What it does | | -------------- | -------------------- | -------------------------------------------------------------------------------- | | claim_work | before editing | records the task + expected files/modules; flags overlaps with other active work | | handoff | when done or blocked | captures what changed, tests run, assumptions, decisions, guardrails | | review_ready | before a PR | records plan, tests, open questions, and provenance |

What you get

SQLite is the local source of truth (gitignored). Concord renders human-readable artifacts you can commit so they show up in PRs:

.concord/
├── concord.db          local source of truth (gitignored)
├── HANDOFF.md          human-readable handoff
├── REVIEW_PACKET.md    review-ready evidence
└── WORK_STATE.json     generated export (optional)

CLI

Agents use the MCP tools; humans use the CLI.

concord init                 # create the .concord/ workspace
concord status               # active work, overlaps, review-ready, open questions
concord tasks                # list all tracked tasks
concord handoff     # print the latest handoff
concord review-packet    # print the latest review packet
concord export markdown      # regenerate .concord/ artifacts
concord doctor               # workspace checks + per-task tool adoption

Try the demo

pnpm demo

Runs the [two-agent overlap demo](./examples/two-agent-overlap/): two agents claim overlapping work (Concord flags it), then one hands off and marks the task review-ready — printing the generated artifacts.

What this is / is not

Shared work-state and guardrails for the coding agents you already use. Not an orchestrator, code reviewer, memory vector DB, or autonomous coding agent.

See also: [Why not just use markdown?](./docs/why-not-markdown.md)

Contributing

See [CONTRIBUTING.md](./CONTRIBUTING.md) and [CLAUDE.md](./CLAUDE.md). This repo is strictly typed (no any, no typecasts), modular, and every PR stays under 600 LOC. Good first issues are labelled good first issue.

License

[MIT](./LICENSE)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Get-Concord-AI
• Source: Get-Concord-AI/concord-mcp
• License: MIT
• Homepage: https://www.npmjs.com/package/@concord-ai/concord-mcp

Install and usage instructions live in the source repository linked above.