agent-mesh lets coding agents talk to each other over MCP
The open source agent-mesh project runs as an MCP server: a connected agent can see neighboring CLIs, open or attach to their sessions, send a prompt, and read the answer back. It claims support for Claude Code, Codex, opencode, Gemini, Grok, and cursor-agent, four of them wired in through ACP, the standard protocol for driving coding agents.
In the project's own tests, codex memorized the key ORCHID-77. Asked the same question, opencode first answered UNKNOWN, then produced the key after pulling it from its neighbor.
Handoff depth is capped by max_ask_depth, 4 by default, with a 300-second timeout per turn. The bridge grants agent permissions itself, so the authors suggest pointing it only at code you're willing to hand to agents. Install on macOS is a single Homebrew command; building from source needs Rust 1.85+. MIT licensed.
Related stories
- Claude Code sessions can now talk to each other
- Omegacode: agent-agnostic workflow implementation
- Claude Code shipped its own TypeScript in npm source maps
- Show HN: Generate a Claude Code or Codex agent fleet from one HTML file
- pairmark races Claude Code against Codex in your repo
- Oh My Subagents keeps Claude Code teams running
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.