Claw-to-Claw (C2C) Protocol

A governed AI agent interoperability protocol — trust, learn, collaborate.

English | 中文

---

What is C2C?

C2C defines how AI agents establish trust, transfer skills, and collaborate — with human and organizational oversight at every step.

Discovery Layer:  A2A    — "What can you do?"
Learning Layer:   C2C    — "Teach me to do it too."   (Sub-Protocol 1)
Collaboration:    C2C    — "Let's work together."     (Sub-Protocol 2)
Execution Layer:  MCP    — "Do this thing for me."

Protocol Structure

C2C v2.0 is organized as a main protocol with sub-protocols:

Component	Purpose	Status
Shared Foundation	Identity, trust, governance, security, encryption	v1.x current
Sub-Protocol 1: Skill Transfer	Teach skills between agents via Knowledge Bottles	v1.x current
Sub-Protocol 2: Trusted Collaboration	Discover and initiate collaboration with trusted partners	v2.0 new

Sub-Protocol 1: Skill Transfer

"Should this agent be allowed to learn this skill?"

Packages, encrypts, and transfers AI skills between agents. Reference implementation: Lobster Distill.

Sub-Protocol 2: Trusted Collaboration

"Which trusted agents can I collaborate with on this task?"

Enables bots to discover availability of trusted partners, select collaborators autonomously, and initiate collaboration — while public nodes only provide visibility, not decisions.

Core principle: Public nodes are pagers, not dispatchers.

Quick Links

Resource	Link
📖 Specification	SPEC.md
📖 规范（中文）	SPEC_CN.md
🦞 Skill Transfer Implementation	Lobster Distill
🌐 Official Site	c2cprotocol.org (coming soon)

Getting Started

Install the Skill Transfer reference implementation:

npx clawhub@latest install qiaoy01/lobster-distill

Contributing

C2C is an open protocol. We welcome:

• 📝 Specification improvements
• 🔧 New sub-protocol implementations
• 🌍 Cross-framework importers/exporters
• 📖 Documentation and translations

License

MIT

---

Claw-to-Claw: Trust, learn, collaborate. 🦞