Home

The open protocol for humans and agents doing accountable work together.

CHAP gives approvals, overrides, handoffs, abstentions and escalations a shared, auditable structure across tools, agent frameworks and organisational boundaries.

MCP connects agents to tools. A2A connects independent agents. CHAP provides the shared collaboration and evidence layer in which humans, agents and services perform, review and govern work together.

This Wiki is CHAP's explanatory layer. The normative specification, schemas and conformance material remain in the main repository.

New to CHAP?

1. Run the five-minute start.
2. See CHAP in twelve practical scenarios.
3. Read how MCP, A2A and CHAP fit together.
4. Explore the implementation registry.

Where does CHAP fit?

• CHAP vs MCP elicitation
• How CHAP works with A2A
• MCP, A2A and CHAP: how the open agent stack fits together
• CHAP vs AG-UI
• CHAP vs workflow engines

How does CHAP capture human judgement?

• CHAP vs application-specific approval logs
• Why AI audit logs do not capture human judgement
• From human override to learning signal
• What CHAP proves, and does not prove, for AI governance

How can I build with CHAP?

• CHAP with LangGraph human-in-the-loop
• Building an independent CHAP implementation

Canonical project resources

• Repository
• Specification
• Five-minute start
• Implementation registry
• Practical scenarios
• Technical report
• Discussions

Project status

CHAP 0.2 is a public draft. TypeScript and Python reference implementations cover the protocol profiles and run against the same JSON-RPC wire and conformance harness. Experimentation, independent implementations and early production feedback are welcome. Deployments requiring strict long-term wire stability should wait for 1.0.

Help shape CHAP

If CHAP addresses a problem you recognise, star the repository, tell us what you are building, or test the protocol from another language or framework.

Last reviewed: 25 August 2026.