Get-Concord-AI/concord-mcp

What it solves

Concord MCP provides a shared communication and coordination layer for AI coding agents operating in the same local repository. It prevents agents from overlapping their work (e.g., editing the same file simultaneously) and allows them to exchange messages, share decisions, and hand off tasks without a human acting as a relay.

How it works

It operates as a Model Context Protocol (MCP) server that maintains a local SQLite database in the repository's .concord/ directory. This database serves as the single source of truth for presence, task memory, and ownership. Agents interact with this state through five core MCP tools: start_work, inspect_work, update_work, transfer_work, and finish_work.

To enable live communication, Concord uses adapters to steer prompts and replies across different agent harnesses (such as Claude Code, Cursor, and Gemini CLI). It also provides a CLI and a TUI dashboard for humans to monitor agent activity and task progress.

Who it’s for

Developers using multiple AI coding agents or different agentic tools within a single codebase who need a way to synchronize their activities and prevent collisions.

Highlights

• Cross-Agent Communication: Enables direct, replyable prompts between agents in different harnesses.
• Collision Detection: Agents can claim files or modules before editing to prevent overlapping work.
• Explicit Handoffs: Provides a structured way to assign, accept, and transfer tasks with evidence and audit history.
• Local-First: All work-state is stored locally in SQLite, ensuring data remains within the repository.
• Broad Compatibility: Supports Claude Code, Codex, Cursor, Gemini CLI, and Grok Build, as well as any other MCP-capable agent.