Skip to content

Summary

Local-first MCP server that lets Claude Code, Codex, Cursor, Gemini CLI and other coding agents message each other, claim files and hand off tasks in one repo.

Features

  • Direct prompts and replies between agents in different coding harnesses
  • File and module claims with overlap detection before edits
  • Versioned assignments and handoffs with an append-only audit history
  • SQLite workspace in .concord/ plus generated handoff and review packets
  • concord check for blocking conflicting edits in hooks or CI

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

  1. Open your agent's MCP configuration file.
  2. Merge the snippet below into it, filling in the environment variables with your own values.
  3. Restart the agent — the "Concord" tools will be available.
{
  "mcpServers": {
    "concord-mcp": {
      "command": "concord-mcp",
      "env": {
        "CONCORD_AGENT_KIND": "cursor",
        "CONCORD_REPO_ROOT": "/absolute/path/to/this/repository"
      }
    }
  }
}

Description

Concord is an open-source coordination layer for AI coding agents working in the same repository. Through one small MCP server, agents in different harnesses (Claude Code, Codex, Cursor, Gemini CLI, Grok Build, Goose, GitHub Copilot CLI, OpenCode, Antigravity and others) can see who else is working, send each other direct prompts and replies, claim files before editing, and hand work over with evidence.

Tools
  • start_work: register presence, claim or accept a task, and report overlapping scope before editing.
  • inspect_work: read workspace and task state, an agent's inbox or outbox, or a prompt/reply thread; flags stale claims.
  • update_work: record task context, or prompt or reply to another reachable agent.
  • transfer_work: assign, accept, decline, release, reassign or offer a handoff, with versioned transitions so two agents cannot both win.
  • finish_work: record evidence and mark a task review-ready, complete or closed.
How it works

State lives in a SQLite database under .concord/ at the repository root (shared across linked Git worktrees), with generated HANDOFF.md, REVIEW_PACKET.md, an event log and a work-state export. A concord CLI reads the same workspace, including a live terminal dashboard and concord check for blocking overlapping edits in pre-commit hooks or CI. Delivery of live messages depends on each harness; concord adapters status reports what reaches whom. An optional Concord Cloud login links agents on different machines.

Setup

npm install -g @concord-ai/concord-mcp, then concord setup in the repository, which registers the server for the detected clients and writes their instructions; a self-contained installer is available for machines without Node. The per-client configs set CONCORD_AGENT_KIND and CONCORD_REPO_ROOT. It is not an orchestrator or autonomous agent. MIT licence.

Related MCP Servers

See all Concord alternatives

MCP: ripwire

by Red Hat

Featured

Red Hat's zero-dependency C++23 code-context engine — ranked call graphs and blast-radius analysis, indexing a repo in under half a second with no server and no database.

Defensive MCP server and CLI that hardens your own release artifacts (JS bundles, Electron ASAR, WASM, SQLite, model files) against AI reverse-engineering agents.

MCP: OmniGet

by tonhowtf

New

A local MCP server that hands coding agents a desktop toolbox: media and torrent downloads, file conversion, dev utilities, linked chats and system cleanup.

MCP: OKF Agent Memory

by okf-memory

Git-native persistent memory for coding agents — sub-300µs BM25 search over version-controlled plain text, with no embeddings, no database and no API costs.

Browse all MCP servers →