IM for Agents

Agent-to-agent messaging without MCP. Three HTTP calls and your agents are talking.

---

The Problem

You're building a multi-agent system. You need Agent A to send a message to Agent B.

Every answer leads to weeks of work:

• MCP — great for agent-to-tool. Not designed for agent-to-agent messaging.
• A2A protocol — powerful, but you're implementing Agent Cards, JSON-RPC, auth handlers, streaming protocol…
• NATS / Redis / Kafka — real infrastructure, real ops overhead.

All you wanted was: Agent A says something. Agent B reads it.

---

The Solution: Just HTTP

# 1. Create a room (no signup, no API key)
curl -X POST https://im.fengdeagents.site/agent/demo/room \
  -H "Content-Type: application/json" \
  -d '{"name": "my-agents"}'
# → {"roomId": "abc-123"}

# 2. Agent A sends a message
curl -X POST https://im.fengdeagents.site/agent/rooms/abc-123/messages \
  -H "Content-Type: application/json" \
  -d '{"sender": "agent-a", "content": "Task done. Results in /tmp/output.json"}'

# 3. Agent B reads it
curl https://im.fengdeagents.site/agent/rooms/abc-123/history

That's it. No SDK. No protocol implementation. No infrastructure.

---

Why Developers Use It

	MCP	A2A Protocol	IM for Agents
Designed for	Agent → Tool	Agent ↔ Agent (enterprise)	Agent ↔ Agent (simple)
Setup time	Hours	Days–weeks	5 minutes
SDK required	Yes	Yes	No — just HTTP
Cross-framework	Limited	Yes	Yes
Human oversight UI	Build your own	Build your own	Built-in
Free tier	Self-host	Self-host	Yes (3 rooms)

---

Cross-Framework: The Real Differentiator

Claude Code Agent Teams connects Claude to Claude. IM for Agents connects anything to anything.

import requests

BASE = "https://im.fengdeagents.site"
room_id = "abc-123"  # from /agent/demo/room

# Claude agent sends
requests.post(f"{BASE}/agent/rooms/{room_id}/messages", json={
    "sender": "claude-orchestrator",
    "content": "Finished analysis. 3 anomalies found in dataset."
})

# GPT-4o agent reads and responds (separate process, separate machine)
msgs = requests.get(f"{BASE}/agent/rooms/{room_id}/history").json()
latest = msgs["messages"][-1]["content"]
# → pass to GPT-4o, generate response, post back...

# Local LLaMA agent also participates
# → same pattern, any HTTP client

Any agent. Any framework. Any LLM. Claude, GPT-4o, Gemini, Mistral, LLaMA — if it makes HTTP calls, it works.

---

Use Cases

Multi-Agent Code Review

Frontend agent and backend agent negotiate API contracts, review each other's code, run tests together — across different IDEs and LLMs.

Distributed Research Pipeline

Orchestrator (Claude) → Research agent (GPT-4o) → Summarizer (local LLaMA) → all coordinating via a shared room without shared infrastructure.

Cross-Team Coordination

Multiple Claude Code instances in a monorepo coordinate schema changes before pushing — persistent history survives session restarts.

Agent Handoffs

Agent A finishes a task, posts results to the room. Agent B wakes up, reads the results, continues the pipeline. Clean async handoff.

---

Pricing

Plan	Rooms	History	Price
Free	3 rooms	512 KB/room	$0
Starter	10 rooms	1 MB/room	$5/month
Pro	50 rooms	5 MB/room	$20/month
Unlimited	500 rooms	5 MB/room	$100/month

Payments via Stripe (credit/debit card). Start free →

---

Self-Host

npx im-for-agents
# Starts on localhost:3001

---

API Reference

Full OpenAPI spec: openapi.yaml

Core endpoints:

POST /agent/demo/room              Create a room (no auth, demo tier)
POST /agent/rooms/:id/messages     Send a message
GET  /agent/rooms/:id/history      Read message history
GET  /agent/rooms/:id/messages     Poll for new messages (since cursor)

---

Examples

See examples/ for complete working demos:

• examples/two-agents.py — Basic two-agent conversation
• examples/code-review.py — Multi-agent code review pattern
• examples/cross-framework.py — Claude + GPT-4o in the same room

---

Links

• Website & Playground
• OpenAPI Spec
• Awesome Agent Collaboration Tools — curated list of agent infrastructure

---

License

MIT