
Agent Comms
Lets LLM agents running in different harnesses talk to each other over a local TCP mesh. You get rooms, DMs, presence, and agent discovery without any filesystem polling or daemon process. Each MCP bridge instance is a mesh peer. The first one up becomes coordinator on port 19876, hands out peer lists, then steps back while agents connect directly. Actions include register, list_agents, create_room, join_room, send, and dm. Messages can hint delivery timing with streamingBehavior: steer for urgent interrupts, followUp for when idle, info for whenever. Works with pi, Claude Code, Codex, or any MCP client. Useful when you want a Claude session to ping a pi session for review, or coordinate work across terminals without shared files.
- GitHub stars
- 37
- tool
- 1
- transport
- STDIO
- last updated
- Nov 11, 2026
Tools
1 toolsPublic tool metadata for what this MCP can expose to an agent.
agent_commsCross-harness communication tool for registering agents, listing agents, creating and joining rooms, sending messages and DMs, reading room history, and updating visibility.
Cross-harness communication tool for registering agents, listing agents, creating and joining rooms, sending messages and DMs, reading room history, and updating visibility.
No parameter schema in public metadata yet.
Sponsors
Readme
Agent Comms
Cross-harness communication mesh for LLM agents: rooms, DMs, presence, and visibility over TCP with zero filesystem dependencies.
Why
Details
- Repository
- github.com/exadev/agent-comms
- Category
- AI & LLM Tools
- Package
agent-comms- Transport
- STDIO
- Registry
- MCP registry, active