trustmodel-agentcert-tag

Verify an AgentCert + TrustScore inside a Python MCP server. Thin, dependency-free client for a TAG verify endpoint — off by default, shadow mode when on (logs, never blocks), safe to merge.

pip install trustmodel-agentcert-tag

from trustmodel_agentcert_tag import verify_gate, extract_token

guard = verify_gate(mode="shadow")            # shadow | enforce

async def handle_tool_call(request):
    token = extract_token(request.headers)    # X-AgentCert-Token (or mTLS-derived)
    await guard(token)                        # logs (shadow) / raises VerifyError (enforce)
    return await dispatch(request)

Enable per environment — nothing runs until you opt in:

The gate never reads request payloads (metadata only) and never makes the allow/deny decision on its own — it returns a structured verdict (VERIFIED / REVOKED / UNVERIFIED / ERROR) your server acts on. MIT licensed. Part of pdxlab/agentcert-tag.

Proof-of-possession (enforce mode)

A presented AgentCert is public, so on a non-mTLS carriage the caller must also send a stapled proof-of-possession (a leaf-signed assertion) in the X-AgentCert-Proof header. The gate extracts it (extract_proof / extractProof) and forwards it with the credential; the verify endpoint rejects a bare cert without it. On mTLS, pass carriage: "mtls" — the handshake already proves possession.