agent-walkie-talkie

MCP Server for autonomous agent-to-agent communication between Claude Code instances.

> Your AI agents talk. You watch on Discord. With popcorn.

---

You're building two projects. Project A needs to integrate with Project B. Sounds simple, right?

Here's what actually happens:

1. You open Project A in Claude Code: *"Hey Claude, I need an API endpoint that sends project data to Project B"*
2. Claude A writes a beautiful API spec
3. You copy the output. All of it. With your mouse. Like it's 2005.
4. You switch to Project B: *"Hey Claude, here's what Project A expects from you..."*
5. You paste the wall of text. Claude B reads it. Asks questions.
6. You copy Claude B's questions. Switch back to Project A. Paste. Wait. Copy the answer. Switch. Paste.
7. Repeat 47 times until both sides agree on a JSON schema.
8. You mass the wrong tab. Lose context. Start over.

By the end, you are the bottleneck. You are the human message bus. You are the walkie-talkie.

What if your agents could just... talk to each other?

---

agent-walkie-talkie is an MCP Server that lets Claude Code instances communicate autonomously through Redis Streams. You start a conversation, they figure it out, you watch the whole thing on Discord (or a web UI) while sipping coffee.

Project A (Claude Code)              Project B (Claude Code)
  "Hey, I need POST /webhook         "Got it. But I need auth.
   with these fields..."               How about X-API-Key header?"
        │                                     │
        └──────► Redis Streams ◄──────────────┘
                      │
                 WalkieTalkie Worker
                 (spawns claude -p for each side)
                      │
              ┌───────┴───────┐
          Discord 🎯       Web UI 🖥️
          #walkie-talkie   localhost:3210
          (thread per       (chat interface
           conversat