Put a signed safety check in front of every tool your agents and developers connect to.
Each tool gets one answer, Safe to connect, Review before you connect, or Do not connect, with a signed 0-100 trust score and an adoption score your CI, gateway or agent can act on.
Public scanning is free forever. Team pilots are free through Q1 2027.
What you can put in place this week.
Gate CI on the answer
Fail the build on Do not connect. Works on GitHub, GitLab, Bitbucket and Azure DevOps.
- uses: AgentAvow/AgentAvow/github-action@main
with:
fail_on: do_not_connectGet an alert when a tool changes
Watch the tools you depend on. An HMAC-signed webhook fires when a score drops or a signed tool definition changes, the rug-pull you'd otherwise miss.
Verify every result offline
Recompute a result byte for byte and check its Ed25519 signature against our public keys. Your auditors don't have to trust us.
Scan private code
Pass a GitHub token for one scan and we never store it. Or run the scanner in your own CI, where your code stays on your runner.
Use the API
The answer, both scores, the findings and the Certified mark come back as JSON, so a policy can require any of them.
Gate agents at runtime
Check each tool before the call with gates for the Vercel AI SDK, Flue, LangChain, Google ADK, Open Policy Agent and LiteLLM.
Straight answers for your review.
Certified. The top mark. It sits beside Safe to connect, and it comes off the moment a check fails. What Certified means →
SOC 2. No SOC 2 yet. Controls shared on request.
Standards. Our tool definition digest is in the TRACE spec. An outside party reproduced our conformance vectors in a Linux Foundation Decentralized Trust lab.
Built on open standards you can check.
You don't have to take a vendor's word for a safety answer. Here is what others can verify.
| What | What it means for your review |
|---|---|
| TRACE spec | Our tool-definition digest is in the TRACE runtime-attestation spec as trace.mcp-tool-definition.v1. Change alerts use the same digest, so a pinned tool can be re-checked by anyone. |
| Linux Foundation Decentralized Trust lab | An outside party reproduced our conformance test vectors byte for byte. The vectors are Apache-2.0. |
| IETF | Our founder co-authors the action-ref Internet-Draft, an individual submission. |
| W3C | Participant in the AI Knowledge Representation and Agent Conformance Community Groups. |
| Signatures | Ed25519 / JWS (RFC 7515) over RFC 8785 canonical JSON. Public keys at a fixed URL. |
| MCP Registry | Our own server is listed in the official MCP Registry. |
Works with the protocols your agents use
MCP
Check live servers by URL, and servers shipped as npm or PyPI packages. Tool definitions are read for poisoning and hidden instructions.
A2A
Import an agent from its A2A agent card into AgentAvow.
x402
26,302 x402 payment endpoints indexed, and a wallet address resolves to the tool it's linked to.
Talk to us.
Tell us what you want to gate. We'll reply with a pilot plan.