Concord MCP
Shared local work-state for coding agents: claim work, hand off, and ship review packets over MCP.
Concord MCP is an open-source MCP server and CLI that gives Claude Code, Codex, Cursor, and other MCP clients a shared per-repo work log backed by local SQLite. Agents claim tasks, append typed progress, leave handoffs, and generate review-ready packets before PRs—without a hosted orchestrator or vector memory store.
Official resources
Selection advice
Quick comparison
Concord MCP focuses on shared task work-state inside one local checkout. waggle standardizes artifact handoff tokens and receipts; Composio and Arcade.dev emphasize SaaS tool catalogs and authorized MCP runtimes.
| Concord MCP | waggle | Composio | |
|---|---|---|---|
| Best for | Coordinating multiple coding agents on the same repo with claim/handoff/review packets | Auditable share-by-reference handoffs of files and plans across agents | Connecting agents to many SaaS apps with managed OAuth |
| Primary surface | Local MCP tools + CLI over a `.concord/` SQLite workspace | Local MCP server + mint/resolve tokens with coverage receipts | Managed toolkits, OAuth, SDK/REST actions, and MCP options |
| Tradeoff | Early local work-state layer—not an orchestrator, hosted sync, or SaaS connector catalog | Strong handoff/reference receipts vs less shared task board semantics | Broad SaaS coverage vs no per-repo agent work log |
Best for
- multi-agent coding on one checkout
- task claim and overlap detection
- pre-PR handoff and review packets
- local-first MCP work-state
Not ideal for
- teams that need a hosted multi-repo sync service
- projects that only need SaaS connector catalogs
- use cases that require a vector memory database
Core concepts
Minimal implementation shape
Install with `npm install -g @concord-ai/concord-mcp`, run `concord install` to write client instructions, `concord init` for a `.concord/` workspace, register the MCP server, then let agents use `claim_work`, `update_task`, and `handoff`.