AgentAvow™Free. No account.
AgentAvow for Claude Code

Claude Code checks your MCP servers and packages before your agent uses them, automatically.

Install the plugin. In your next session, Claude tells you which ones read Safe to connect, Review before you connect, or Do not connect, and why.

1/plugin marketplace add AgentAvow/AgentAvow
2/plugin install agentavow-trust@agentavow

Run each line in Claude Code, then start a new session. Or open /plugin and find AgentAvow Trust under Discover.

Free, no accountThe plugin is MIT licensedOnly the server URL or package name leaves your machine

Your first session opens with one line.

AgentAvow pre-check: checked 8 MCP servers. 5 Safe, 2 Review, 1 Do not connect.
Needs attention: 'example-server' Do not connect, one critical finding.
Dependencies: checked 12 of 38. 11 Safe, 1 Review.

It stays quiet until it matters.

Safe to connectStays silent. A tool that cleared the full gate reads Safe to connect · Certified.
Review before you connectOne line with the reason. The install goes ahead.
Do not connectBefore an add or install, Claude Code asks you first. You decide.
Tool definition changedClaude Code asks before the call.

Each result carries a signed 0-100 trust score and an adoption score, with a link to the full report.

A server you approved can change later.

In a rug-pull, a tool passes review, then its definition changes to do something else. The plugin compares what a remote server serves now with the signed definition from the check. If they differ, Claude Code asks you first.

signed sha256:811d1414506d36dccc1a273a29714fe29d1d2f7f2220dc16b61fec0f972d70da
served sha256:b2163bf3f4aa19a65db1fc25b9bcf0609d8116603226490a708f2d7860be174f

From our rug-pull demo ↗, run on localhost with test data.

Why developers can trust the answer.

Hundreds of Claude Code machines connect each day.

Each result is signed with Ed25519, and you can verify it offline against our public keys.

Our tool definition digest is in the TRACE spec as trace.mcp-tool-definition.v1.

An outside party reproduced our conformance vectors byte for byte in a Linux Foundation Decentralized Trust lab.

Our founder co-authors the action-ref Internet-Draft, an individual submission to the IETF.

What it checks

  • MCP servers, at session start and on claude mcp add
  • Packages Claude installs by name with npm, pnpm, yarn, bun, pip, uv, pipx or poetry
  • Your project's direct dependencies, up to 15 per session start

What it doesn't do

  • Lockfile installs, requirements files and URL installs are left alone
  • If AgentAvow can't be reached, your work carries on with no check
  • A check can miss things. It warns, and you decide

Add it now.

1/plugin marketplace add AgentAvow/AgentAvow
2/plugin install agentavow-trust@agentavow

Needs Claude Code and python3. Remove it any time from /plugin.

What this page measures: page views and clicks on the copy buttons, with the campaign tag from your link and a random ID stored in your browser. This page also loads the Meta Pixel, so we can see how many people our Meta ads bring here and how many copy the install command. It tells Meta about the page view and the first copy, and Meta may set a cookie. No other page on agentavow.com loads it, and it stays off if your browser sends Global Privacy Control or Do Not Track. Details and how to opt out are in our privacy policy.

PrivacyTerms