coord: The Dev-Loop Coordinator for AI Agents

coord serves as a powerful local coordinator designed specifically for parallel AI coding agents, supporting multiple environments seamlessly. By acting as a unified interface for various coding tasks across platforms like Claude Code, Cursor, and Codex, it enables collaborative coding without the agents being aware of each other’s activities.

Key Features

• Atomic Task Claims: Utilizing atomic claims, coord ensures that multiple agents do not attempt to claim the same task simultaneously; only one agent succeeds in claiming a task, leading to clear accountability and resolution of tasks.

• Blocking Watch Primitive: The coord wait command allows agents to block execution until certain tasks are completed. This approach enhances efficiency by eliminating unnecessary polling efforts, as agents can simply wait for critical tasks.

  coord wait --kind ack --name-contains 'v1.2'
  

• Two-in-One Protocols: coord operates with two protocols while maintaining a single daemon. It communicates with A2A on its HTTP surface and functions as an MCP server for integration with IDE clients that do not directly support A2A.

• Markdown Audit Trail: Each state change within the system generates a markdown note with wikilinks to other related tasks. This feature can be exported to Obsidian, providing a clear graphical representation of task relationships and actions over time.

• Simple Deployment: With coord, everything can be managed through a single binary executable, significantly streamlining project setup and usage without the complexities of additional environments such as Python virtual environments or Docker.

Architecture Overview

The architecture integrates easily with various agents thanks to its robust communication model. Multiple agents can connect over loopback HTTP, with SQLite managing data access effectively to allow concurrent interactions without performance degradation.

Example Usage

To quickly get started with coord:

coord serve --vault ~/coord-vault  # Start the daemon
coord init                         # Initialize project files
coord top                          # Monitor task activities

Why Choose coord?

In contrast to other local coordination layers that might offer a more complex mailbox or messaging system, coord focuses on efficient claim-and-watch semantics tailored to enhance coding workflows across multiple AI platforms. Its lightweight architecture and straightforward operational model make it a valuable tool for developers aiming for seamless collaboration with AI coding agents.