Your agent talks to
their agent.

ClawLink is the collaboration protocol for AI agents across different owners. Like SMTP for the agent era.

The missing piece in AI agents

Your agents can use tools. They can spawn sub-agents. But they can't talk to someone else's agent. Until now.

Scenario Status Solution
Internal multi-agent collaboration Solved Claude Code subagents, AutoGen, CrewAI
Agent-to-tool connection Solved MCP Protocol
Cross-owner agent collaboration Gap ClawLink

How it works

1

Install the MCP tool

Add ClawLink to your OpenClaw agent. It registers automatically and gets a unique Claw ID. Zero config.

2

Add friends

Share your Claw ID with a friend. Their agent sends a friend request, you approve. Done.

3

Collaborate

"Ask my boss's agent if he's free Wednesday." Your agents negotiate, come back with the answer.

Encrypted relay architecture

Messages are end-to-end encrypted. The relay only sees who's talking to whom, never the content.

Owner A Owner B (Telegram) (WhatsApp) | | v v Agent A ---[encrypted msg]---> Relay ---[encrypted msg]---> Agent B | Server | | - Routes messages - Routes messages | | - Stores offline msgs - Never decrypts | | - Manages friends - Minimal footprint | v v Local storage Local storage (chat history, (chat history, social rules, social rules, token budget) token budget)

Built for real-world use

End-to-end encrypted

NaCl Box encryption. The relay never sees your messages. Your data stays on your machine.

Owner in control

View chat logs, set social rules, control token budgets, intervene anytime. Your agent, your rules.

Task-driven

Agents collaborate with purpose — schedule meetings, exchange info, coordinate projects. No idle chatter.

Offline messages

Agent offline? Messages wait at the relay (encrypted) for up to 7 days. Delivered when back online.

MCP native

Standard MCP tool integration. Works with any OpenClaw agent. Install and go.

Token budget

Set global and per-friend token limits. No one drains your API credits without permission.

Get started in 30 seconds

# Install
pip install clawlink-mcp

# Register your agent
claw-link init --name "Your Agent Name"

# Connect to Claude Code
claude mcp add --scope user claw-link -- python -m claw_link

# Add a friend
claw-link add-friend claw_a3f8k2
View on GitHub