Part of the security infrastructure for AI agents

Verify trust before you install.

Look up community trust data for MCP servers, A2A agents, and AI tools. Trust level, recent scan results, advisories, and dependency risk in one query.

$npx ai-trust check server-filesystem

One package. One command. No signup. Runs locally.

View on GitHubRead the docs

What you get back.

One query. Resolved name, trust level, scan recency, advisories, and dependency roll up. No score theater. Every signal traces back to a registry record or a recent scan.

Terminal
$ npx ai-trust check server-filesystem
ai-trust v0.7.0 ยท resolved: @modelcontextprotocol/server-filesystem
Trust
Level 4.Verified
Pass --json for machine readable output.
Signals
Publisher verified
Last scan 2026-04-22 . hackmyagent v0.23.0
Findings 0 critical, 0 high
Advisories none open
Dependencies 3 . all scanned
Source npm

Three commands.

One package, a whole manifest, or a list of names. Same trust graph behind every call.

ai-trust check

Look up a single package

Returns trust level, last scan date, advisories, and dependency risk. Pass --scan-if-missing to scan locally with HackMyAgent when the package is not yet on file.

$ ai-trust check server-filesystem

ai-trust audit

Audit a dependency file

Parses package.json, requirements.txt, or any supported manifest and queries every dependency in one pass. Use --min-trust to fail the run when any package falls below a threshold.

$ ai-trust audit package.json --min-trust 3

ai-trust batch

Verify many packages at once

Pass a list of names. Useful for ad hoc allowlists, supply chain reviews, and CI gates that read package names from a file.

$ ai-trust batch express lodash chalk --min-trust 2

Five trust levels.

Every package resolves to a level from 0 to 4. Set --min-trust in CI to fail the run when any dependency falls below your threshold.

0
Blocked
Known malicious or actively exploited. Do not install.
1
Warning
Open advisories or scan failures. Review before install.
2
Listed
Discovered in the registry but not yet scanned.
3
Scanned
Recent HackMyAgent scan on record. Findings visible.
4
Verified
Publisher verified plus passing scan plus community signals.

What trust is built from.

Eight signal classes. Every level resolves from registry records, recent scans, and anonymized community reports. No invented numbers.

Scan history

What HackMyAgent has found across recent scans of this package.

  • Static, semantic, and adversarial findings
  • Severity counts and check IDs
  • Most recent scan timestamp

Publisher provenance

Who shipped the package and whether the registry trusts them.

  • npm and PyPI publisher identity
  • Verified publisher signals
  • Domain and org continuity

Open advisories

Known vulnerabilities and security advisories tied to this package or its dependencies.

  • GitHub advisory database
  • CVE references where available
  • Severity at time of query

Dependency risk

Trust roll up across the dependency tree, not just the surface package.

  • Transitive dependency review
  • Lowest trust level in the tree
  • Unscanned dependency callouts

Community signals

What the community has reported back about this package.

  • Anonymized scan contributions
  • Reported issues and false positives
  • Usage breadth across registries

Drift and freshness

Whether the trust signal is current or stale.

  • Time since last scan
  • Version drift between releases
  • Scan refresh recommendations

Activity and adoption

Lifecycle signals for the package itself.

  • Release cadence
  • Maintenance status
  • Reach across npm, PyPI, and Docker

Source registry

Where the artifact actually lives and how to reach it.

  • npm, PyPI, Docker, GitHub source
  • Canonical name resolution
  • Direct registry links

Example queries.

Lookups, audits, and on demand scans across npm, PyPI, and MCP package names.

$ ai-trust check @modelcontextprotocol/server-filesystem

MCP server lookup with full resolved name

$ ai-trust check server-filesystem

Short name resolves to canonical package

$ ai-trust audit package.json --min-trust 3

Fail CI when any dep is below trust level 3

$ ai-trust check mcp-server-xyz --scan-if-missing

Scan locally when the package is not yet on file

$ ai-trust audit requirements.txt --scan-missing --contribute

Audit, scan unknowns, contribute anonymized results

$ ai-trust batch express lodash chalk

Bulk verify a list of names

Scan on demand.

Package not yet on file? Add --scan-if-missing and ai-trust downloads the artifact and runs a HackMyAgent scan locally. Add --contribute to share anonymized findings back to the community.

Terminal
# Scan a package not yet on file
$ ai-trust check mcp-server-xyz --scan-if-missing

# Scan and contribute results back to the community
$ ai-trust check mcp-server-xyz --scan-if-missing --contribute

# Audit a manifest, scan unknowns, contribute results
$ ai-trust audit package.json --scan-missing --contribute

Community trust, opt in.

Share anonymized scan results to help the community identify unsafe packages. No personal data, no source code. Check IDs, pass and fail counts, severity. That is the entire payload.

Asked once

On first scan, ai-trust prompts and stores your choice in ~/.opena2a/config.json. Change it any time.

Privacy first

Only check IDs, pass and fail counts, and severity are sent. No file paths, no descriptions, no fix text, no code.

Shared across tools

Opt in carries across opena2a-cli, hackmyagent, and ai-trust. Configure once, contribute from any tool.

CI integration.

Every command supports --json. Non zero exit when packages are blocked or fall below your trust threshold.

.github/workflows/trust-gate.yml
- name: Verify AI package trust
  run: npx ai-trust audit package.json --min-trust 3 --json

- name: Check MCP server, scan if missing
  run: npx ai-trust check @org/mcp-server --scan-if-missing --json

- name: Audit with on demand scanning for unknown deps
  run: npx ai-trust audit package.json --scan-missing --contribute

How it fits the ecosystem.

ai-trust queries the trust graph that HackMyAgent scans populate. Same data is also reachable through the OpenA2A CLI when you want one tool for everything.

HackMyAgent populates

Every HackMyAgent scan that opts in feeds anonymized findings into the trust graph.

ai-trust queries

Lookup, audit, and batch verify against the same trust graph before you install.

OpenA2A CLI delegates

opena2a trust and opena2a registry route through ai-trust under the hood.

Verify before you install.

One command. No signup. Runs locally.

$npx ai-trust check server-filesystem
Star on GitHubAll tools