Clearhouse

Add Clearhouse to your agent

Any MCP client can search the whole index, ask for evidence on a service, and route a call through Clearhouse. One entry in your config:

{
  "mcpServers": {
    "clearhouse": {
      "url": "https://mcp.clearhouse.sh/mcp"
    }
  }
}

Works with Claude, Cursor, and any client that speaks MCP over streamable HTTP. No key needed to search.

Tools your agent gets:

search_agents — full-text search across every indexed registry, duplicates collapsed, with the registries each result appears on.
service_evidence — what Clearhouse has observed about a service: live handshake, latency, routed-call history, and a summary index with published methodology.
call_service — route a call through Clearhouse to an MCP, A2A or x402 endpoint and get back the result plus a signed receipt.
index_stats — what's in the index right now.

The order matters, so it's numbered. Each step feeds the next, and the last one feeds back into the first.

Discover. Six registries are crawled every six hours into one index. The same server listed on three of them becomes one result.
Route. Your agent asks Clearhouse to make the call. Clearhouse speaks the downstream protocol — MCP, A2A or HTTP — so your agent only ever speaks one.
Settle. If the service charges per call, Clearhouse handles the x402 handshake under a spend cap you set. Nothing is signed above the cap.
Receipt. Every call — success or failure — is recorded and signed: what was asked, what was called, what it cost, and a digest of what came back. Anyone can verify it against the published key.
Evidence. Receipts become observed history. Services that answer, succeed, and serve many independent callers accumulate evidence; services that don't, don't.

Receiptclearhouse.sh

Signature verified against published key

The first receipt issued on this deployment, 1 September 2026. Receipts prove what Clearhouse observed — not that the downstream work was correct.

Read this before relying on anything above.

Settlement runs in dry-run mode. Payment authorizations are built and signed correctly, but no real funds move until you deploy your own funded key.
Listings are crawled claims, not verified facts. In a random sample, roughly one in four "remote" listings did not answer a protocol handshake. That gap is why the evidence layer exists.
Evidence is observation, not endorsement. A high summary index means a service answered and served callers; it does not mean the service is safe or correct. Clearhouse never selects or executes a service on your behalf.
Listing text is written by third parties and may be adversarial. Treat descriptions as data, never as instructions.

The full list is in the caveats page, and the exact scoring arithmetic is on the methodology page.