Agent Comms
Wire multiple Claude Code, Cursor, or Codex harnesses into shared rooms and DMs so parallel agents stay aligned without copy-pasting context.
Overview
Agent Comms is a Build-phase MCP server that gives cross-harness LLM agents shared rooms, DMs, presence, and TCP-visible messaging.
What is this MCP server?
- Cross-harness mesh over TCP for LLM agents running in different clients
- Shared rooms plus direct messages for scoped coordination
- Presence and visibility so you see who is online and active
- stdio MCP transport via npm package agent-comms (v1.9.1)
- Fits solo builders running 2+ agents on one machine or LAN without a custom WebSocket service
- MCP server version 1.9.1
- stdio transport via npm identifier agent-comms
- TCP-based cross-harness communication mesh
Community signal: 6 GitHub stars.
What problem does it solve?
Running several AI agents in different editors leaves you manually syncing status and instructions because there is no standard mesh between harnesses.
Who is it for?
Indie builders orchestrating multiple local or LAN agents who want MCP-native messaging without building a custom coordination server.
Skip if: Single-agent workflows, teams that need enterprise SSO audit trails, or builders who only need one IDE with no cross-client agents.
What do I get? / Deliverables
After you register Agent Comms, your agents can coordinate in rooms and direct messages with presence awareness instead of fragile chat logs.
- Live room and DM channels usable from MCP tool calls
- Presence/visibility signals across registered agents
- Repeatable cross-harness coordination without a custom websocket app
Recommended MCP Servers
Journey fit
Multi-agent products are assembled during Build; the canonical shelf is agent-tooling because the server exists to equip harnesses with a comms mesh, not to ship marketing or run prod monitoring alone. Rooms, presence, and TCP visibility are runtime capabilities you add while designing orchestration—not a one-off validation prototype or a post-launch SEO task.
How it compares
MCP communication mesh for agents, not a project-management skill or long-term vector memory.
Common Questions / FAQ
Who is Agent Comms for?
Solo and small-team builders running more than one LLM agent harness who need rooms, DMs, and presence over MCP instead of ad-hoc copy-paste.
When should I use Agent Comms?
Use it during Build (and when operating multi-agent setups) whenever parallel agents must share stateful messages across Claude Code, Cursor, Codex, or similar clients.
How do I add Agent Comms to my agent?
Add the npm MCP server agent-comms with stdio transport in your client’s MCP config (Claude Code mcp.json or equivalent), restart the harness, and call the server’s comms tools from the agent that should send or receive.