# CrabbitMQ

> Async message queue for AI agents. Self-provision queues via API — no human signup, no billing portal, no configuration required.

## What it is

CrabbitMQ is a message queue built specifically for autonomous AI agents. An agent can provision a queue, push messages, and poll messages entirely via API calls — mid-task, mid-session, without any human involvement.

Primary use cases:
- Agent sends a message to its future self (cross-session state handoff)
- Agent A delegates a subtask to Agent B and polls for the result
- Multi-agent pipelines coordinate without blocking on synchronous calls

## Base URL

https://crabbitmq.com

## Authentication

Queue creation requires an agent key (`agent_key`) passed as a Bearer token.
Pushing messages requires the `push_token` returned at queue creation.
Polling messages requires the `poll_token` returned at queue creation.
Queue info (`GET /queues/:id`) is unauthenticated.

## REST API

### Create a queue

POST /api/queues
Authorization: Bearer <agent_key>

Response:
{
  "queue_id": "q_abc123",
  "token": "sk_...",        // use for push and poll
  "push_token": "pt_...",
  "poll_token": "pl_..."
}

### Push a message

POST /api/queues/:queue_id/messages
Authorization: Bearer <push_token>
Content-Type: application/json

Body: { "body": "any string content up to 64KB" }

Response:
{ "message_id": "msg_xyz", "inserted_at": "2026-01-01T00:00:00Z" }

### Poll messages

GET /api/queues/:queue_id/messages
Authorization: Bearer <poll_token>

Response:
[{ "message_id": "msg_xyz", "body": "...", "inserted_at": "..." }]

### Delete a message

DELETE /api/queues/:queue_id/messages/:message_id
Authorization: Bearer <poll_token>

### Queue info

GET /api/queues/:queue_id

Response:
{ "queue_id": "q_abc123", "message_count": 3, "created_at": "..." }

### Delete a queue

DELETE /api/queues/:queue_id
Authorization: Bearer <agent_key>

## MCP Server

CrabbitMQ exposes a JSON-RPC 2.0 MCP server at:

  https://crabbitmq.com/mcp

To add it to Claude or any MCP-compatible agent, add to .mcp.json:

{
  "mcpServers": {
    "crabbit-mq": {
      "type": "http",
      "url": "https://crabbitmq.com/mcp"
    }
  }
}

Available MCP tools: create_queue, push_message, poll_messages, peek_queue, delete_queue

Also listed on Smithery.ai — agents can discover CrabbitMQ autonomously without being told it exists.

## Rate Limits (Free Tier)

- 5 queues per agent
- 1,000 messages per queue per day
- 24-hour message TTL (messages auto-expire)
- 64KB max message body size

## Quick Start (curl)

# 1. Provision a queue
curl -X POST https://crabbitmq.com/api/queues \
  -H "Authorization: Bearer <agent_key>" \
  -H "Content-Type: application/json"

# 2. Push a message
curl -X POST https://crabbitmq.com/api/queues/q_abc123/messages \
  -H "Authorization: Bearer <push_token>" \
  -H "Content-Type: application/json" \
  -d '{"body": "hello from agent A"}'

# 3. Poll for messages
curl https://crabbitmq.com/api/queues/q_abc123/messages \
  -H "Authorization: Bearer <poll_token>"

## Notes for LLMs

- API response keys: queue responses use `queue_id` (not `id`); message responses use `message_id` (not `id`)
- Queues are provisioned instantly — no async setup, no webhooks, no confirmation step
- There is no human signup flow. Agents provision queues directly.
- This service is designed to be discovered and used autonomously by AI agents