The agentic web
CHP is the evidence layer of the agentic web.
The agentic web is converging on standards for discovery (llms.txt, capabilities.txt), invocation (MCP), and identity (Web Bot Auth). One layer is still unsolved: evidence — what an agent actually did, whether it was allowed, and whether you can prove it later. CHP is that layer.
Identity answers who an agent is. CHP answers what it did, whether it was allowed, and proves it — governed, tamper-evident, replayable execution at the capability boundary.
The stack
Four questions, four layers.
Discovery
robots.txt · sitemap · llms.txt · capabilities.txt
What can this host do?
CHP contributes capabilities.txt (governance-aware discovery).
Invocation
MCP · HTTP · A2A
How do I call it?
CHP runs a live MCP server — agents query and invoke it directly.
Identity
Web Bot Auth (cryptographic agent identity)
Who is this agent?
CHP aligns with it — identity is the input to accountable evidence.
Evidence
CHP
What did it do? Was it allowed? Can I prove it?
The layer CHP owns: governed, provable, replayable execution.
Discovery → invocation → identity → evidence. The first three are maturing fast (MCP is near-universal; Web Bot Auth went to production in 2026). The hardest question — what happened, and can you prove it? — is the one CHP answers.
Why it is the hard layer
Evidence has to outlive the agent that produced it.
When an agent acts in consequential work, the question arrives later and from a skeptic — a security reviewer, an auditor, a regulator. A log the framework wrote, in its format, at its discretion, is exactly the self-attestation a skeptic discounts. Evidence must be portable, structured, and verifiable on its own terms — and that is a property of a shared protocol, not a vendor feature.
evidence chain · tamper-evident
each block hashes the one before it — chain verifies ✓
tip: click a block to alter it
Each event hash-chains to the prior one — tamper-evidence you can verify. Try it.
We practice it
CHP is built to be consumed by agents.
A protocol for governing agent actions should be exemplary at being discovered and used by agents. Connect an agent to CHP's MCP server and it can learn the protocol, browse the capability catalog, and adopt it — and every call comes back wrapped as hash-chained CHP evidence. The protocol demonstrating itself.
FAQ
Complementary, not competing.
Does CHP replace MCP?
No — they are complementary layers. MCP answers "what can the model call" (invocation). CHP answers "what actually happened, who was denied, and can I replay it" (evidence). An MCP server can be wrapped as a CHP host so its tool calls emit evidence; a CHP capability can be exposed through an MCP tool surface. CHP even runs its own MCP server.
Does CHP replace Web Bot Auth?
No. Web Bot Auth (Cloudflare/IETF) cryptographically proves WHO an agent is. It does not record WHAT the agent did or whether it was allowed. CHP is that remainder. The two compose: identity in, accountable evidence out.
Is CHP itself agent-ready?
Yes. CHP publishes an A2A agent card at /.well-known/agent-card.json, an MCP server-card, capabilities.txt, and a live MCP server at /api/mcp that any agent can connect to and learn from — every call dogfooded as hash-chained CHP evidence.
How does capabilities.txt relate to the A2A agent card?
They are complementary discovery layers, and CHP publishes both. capabilities.txt is the static, crawlable, governance- and evidence-aware advertisement; the A2A agent card is the recognized capability manifest agents look for at a well-known path.
Build on the evidence layer.
Start with agents — capture exactly what yours did in one command — or build a vertical with us.