MCP/Concord MCP
Concord MCP logo

Concord MCP

Verified Blue (500+ GitHub stars)

Live messaging, task coordination, file claim overlap detection, and review handoffs for coding agents across Claude Code, Codex, and Cursor.

developer-tools•Developer Tooling & Discovery
GitHub668
License:MIT
VERIFIED
ByConcord AI@Concord AI

About this Server

Concord MCP is the open-source, local-first communication and coordination layer for AI coding agents. It connects Claude Code, Codex, Cursor, Gemini CLI, Grok Build, Goose, and GitHub Copilot CLI in a shared workspace backed by local SQLite. Agents can send live messages across different harnesses, claim files and detect overlap conflicts before editing, preserve task memory, and hand off tasks with complete review evidence.

Real-time Interaction

Direct low-latency RPC protocol communication with structured JSON schemas.

Scoped Sandboxing

Restricted capability bounds to prevent accidental modifications or leakage.

Security & Best Practices

Local-First SQLite Storage: Stores work state inside .concord/concord.db within the repo root with zero external cloud dependencies.
Root Directory Scoping: Confines file claiming and task resolution strictly within CONCORD_REPO_ROOT.
Monotonic Version Locking: Enforces atomic state transitions using version checks to prevent race conditions.

When to Use (and When Not To)

Best For
  • •Running multi-agent coding workflows where Claude Code, Codex, and Cursor work simultaneously in one repo
  • •Preventing conflicting code modifications on the same files through preemptive claims
  • •Generating verified handoff review packets with scope, risks, and provenance before PR creation
Not For / Limitations
  • •External Web3 RPC endpoint discovery and multi-chain comparison (use Chain.Love Gateway instead)
  • •Customer-facing Communicate workspace campaigns (use Communicate MCP instead)
  • •Apple Developer Swift and HIG documentation lookup (use Apple Developer Docs MCP instead)

Install & Setup

terminal command
$claude mcp add concord -- concord-mcp

Exposed Tools & Endpoints

5 tools

Functions your AI agent can invoke via this Model Context Protocol server

start_work

Registers agent presence, claims or accepts a task, and detects potential file/module scope overlaps before edits.

inspect_work

Reads workspace and task state, checks agent inbox/outbox, and inspects active prompt/reply threads.

update_work

Records task context or delivers live prompt/reply messages to another active workspace agent.

transfer_work

Assigns, accepts, releases, reassigns, or offers handoffs of version-controlled work packages.

finish_work

Records test evidence and marks tasks review-ready, complete, or closed.