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 ```bash 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: ```text .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. ```bash concord init # create the .concord/ workspace concord status # active work, overlaps, review-ready, open questions concord tasks # list all tracked tasks concord handoff <task-id> # print the latest handoff concord review-packet <id> # print the latest review packet concord export markdown # regenerate .concord/ artifacts concord doctor # workspace checks + per-task tool adoption ``` ## Try the demo ```bash 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`](https://github.com/Get-Concord-AI/concord-mcp/labels/good%20first%20issue). ## License [MIT](./LICENSE)