AgentTrust (getagenttrust.com)

npx add-mcp 'https://getagenttrust.com/api/mcp'

Know whether to trust an agent before you call it.

AgentTrust checks an agent endpoint against the monitoring, reliability and ownership evidence it has already collected, and returns a machine-readable trust decision — without contacting the endpoint.

No endpoint handy? Check AgentTrust's own test agent

Example Agent

https://agent.example.com/a2a

Recommended

Confidence: High

Health

Healthy

Reliability

96/100 · Excellent

Evidence freshness

Current

Endpoint ownership

Verified

No caveats reported

Illustrative data for a fictional agent — not a live lookup.

Discover → Trust → Invoke

AgentTrust sits between finding an agent and calling it: an endpoint goes in, observed evidence is read, and a trustDecision comes out.

1. 01

   Discover

   Your agent finds another agent's endpoint — in an A2A Agent Card, a registry, or its own configuration. endpointUrl
2. 02

   Trust

   Before calling it, your agent asks AgentTrust. It reads the evidence AgentTrust has already observed for that exact URL — it never contacts the endpoint.
   • Health
   • Reliability
   • Evidence freshness
   • Endpoint ownership check_agent_trust({ endpointUrl }) trustDecision
3. 03

   Invoke

   Proceed on a recommendation, or apply your own policy to the confidence and reasons. The decision stays yours. trustDecision.recommended

Observed evidence, one deterministic rule.

// recommended

status is healthy AND a current score ≥ 50

// confidence

• high — verified and score ≥ 90
• medium — verified and score ≥ 50
• low — not verified, or score below 50
• insufficient_data — no current score

Not a community rating and not a security guarantee — AgentTrust reports what it has observed so you can decide.

Health status

Healthy, degraded or down — derived from consecutive results of scheduled health checks, or heartbeats for push-mode agents.

Reliability score

0–100 from the last 7 days of checks: uptime, latency, consistency and incidents. Computed once there are at least 5 checks.

Evidence freshness

Current when the agent's latest check produced it — which takes at least 5 checks in the 7 days before it — and it's at most 50 hours old. An out-of-date score is shown for reference but never counts toward a recommendation.

Endpoint ownership

The owner publishes a token at /.well-known/agenttrust-verification.txt on the endpoint's origin. Optional — it raises confidence, but is never required for a recommendation.

The same check, over MCP.

Any MCP client can call check_agent_trust before invoking an unknown agent.

• Read-only — reads AgentTrust's existing observations
• No AgentTrust account or API key needed for check_agent_trust
• Exact match on the endpoint URL you pass
• Machine-readable trustDecision: recommended, confidence, reasons
• Never contacts the endpoint being checked
• Anonymous calls are rate-limited per caller IP

https://getagenttrust.com/api/mcp

Copy

Tool call

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "check_agent_trust",
    "arguments": {
      "endpointUrl": "https://agent.example.com/a2a"
    }
  }
}

Result — structuredContent, example

{
  "matched": true,
  "slug": "example-agent",
  "name": "Example Agent",
  "status": "healthy",
  "verified": true,
  "reliabilityScore": 96,
  "reliabilityScoreStatus": "fresh",
  "trustDecision": {
    "recommended": true,
    "confidence": "high",
    "reasons": []
  }
}

Check an agent before you call it.

Paste its exact endpoint URL. Running your own agent? Register it so others can check it too.