Vouch
8 / 10 endpoints operational

The trust layer for the agentic web.

Choosing between MCP, A2A, and REST endpoints shouldn't be guesswork. Vouch surfaces real-world community adoption, tests protocol conformance, and publishes an independent trust score — so you can select on evidence.

Browse the registryRegister an endpoint
1 min  probe interval·MCP · A2A · REST  protocols·No login  all data public·/api/v1/status  machine-readable
8 of 10 operational
10 endpoints monitored
90 days · oldest firstUP · DEGRADED · DOWNView full monitor →
01 · The problemRegistries list. Vouch verifies.

A listing is not a guarantee.

→ 01

Being listed says nothing about working correctly.

An endpoint in a registry tells you it exists. It says nothing about whether the MCP handshake completes, whether the schema matches the docs, whether it was up last week, or whether any real agent has ever called it successfully. Vouch answers all four.

→ 02

Self-reported data has a conflict of interest.

Status pages are written by vendors. Vouch is not. We probe from outside, publish every result in the open, and have no commercial incentive to soften a bad reading. Independent verification is structurally more trustworthy than self-reporting — the same reason companies get SOC2 audits.

→ 03

Developers and agents need the same signal.

A developer choosing between seven MCP servers needs evidence at integration time. An agent routing tool calls needs the same evidence at runtime. Vouch publishes one source of truth — over the web, over REST, over MCP and A2A — so both get it without scraping.

02 · Who it's forOne source of truth, three use cases
For developers

Select on evidence.

Search the registry, inspect the Class rating, review conformance scores and 30-day reliability history. Make the integration decision on data — not docs, not reputation, not GitHub stars.

class A · Vouched
uptime_30d 100.0%
conformance pass
agents_vouched 1,240
For endpoint owners

Earn independent certification.

Register in 60 seconds. Vouch probes your endpoint continuously, scores it independently, and publishes the result. The badge is a third-party receipt — not a self-report.

vouch-agent.com/your-slug/badge
✓ VOUCHEDCLASS A
For agents

Route on live trust data.

The same data that powers this site is served over MCP and A2A. Agents can check status, compare endpoints, and fail gracefully on degraded services — in real time, programmatically.

tool vouch.lookup
↳ endpoint "mcp.linear.app"
↳ status "operational"
↳ class "A"
03 · The rating systemFour classes. Two pillars. One independent score.

Not just up. Actually verified.

A server can return HTTP 200 and still be non-conformant, poorly implemented, or unused by any real agent. The Class rating separates those cases clearly.

D
Reachable
HTTP 200 — endpoint responds
SyntheticCommunity
C
Conformant
MCP handshake completes · schema validates
SyntheticCommunity
B
Monitored
30-day synthetic uptime ≥ 99%
SyntheticCommunity
A
Vouched
Class B + independently verified by 500+ agents
SyntheticCommunity
04 · What we testProtocol-aware probing, not just pings
MCP

Model Context Protocol servers. Probed for transport health, initialize handshake completion, schema conformance, and tool/resource listing validity.

  • Handshakeinitialize · initialized
  • Conformanceschema-validated response
  • Latencyp50 · p95 tracked
A2A

Agent-to-Agent protocol endpoints. Probed for agent card structure, capability advertisement correctness, and message round-trip health.

  • Agent card/.well-known/agent.json
  • Structurename · url · skills validated
  • Latencyp50 · p95 tracked
REST

HTTP endpoints. Probed with configurable expectations: status code, response time budget, and optional body schema match.

  • MethodGET · POST · HEAD
  • Expectstatus · latency · schema
  • Latencyp50 · p95 tracked
05 · The dataPublic, structured, machine-readable

The page is the API.

Every status, score, and latency number you see on this site is published as structured JSON at a stable URL. No private dashboards. No editorial layer. What we measure is what we publish.

REST
GET /api/v1/status
MCP
vouch.lookup · vouch.history
A2A
vouch.agent / tasks/send
// GET /api/v1/status
{
  "endpoints": [{
    "slug": "agentic-shelf",
    "protocol": "MCP",
    "class": "A",
    "status": "operational",
    "uptime_30d": 1.0,
    "conformance": "pass",
    "agents_vouched": 1240
  }]
}
06 · How it worksRegister once. Scored continuously.
STEP 01

Register.

Submit via form or POST /api/v1/register. No account required. Accepted endpoints begin probing within minutes.

STEP 02

Probe.

Synthetic checks run every minute. Protocol conformance and schema validation run on every check. Results are written immediately.

STEP 03

Classify.

Each endpoint is scored Class D through A based on conformance, reliability history, and independently observed community adoption.

STEP 04

Publish.

All data is public, machine-readable, and served over REST, MCP, and A2A. The score updates the minute the data changes.

07 · ArchitectureTwo sources. One engine. One score.

Independent data, combined into one score.

Community telemetry gives Vouch a real-world adoption signal — the one no synthetic tool can replicate. Probing gives it a controlled, independent conformance baseline. Neither alone is sufficient. Combined, they produce a rating that is both practically validated and technically rigorous.

Community alone→ real adoption, no conformance proofSynthetic alone→ technically compliant, no usage proofBoth combined→ Class A · Vouched
Pillar 1Community-observed
Community SDK
  • Agents observe real tool calls
  • Batched + signed, never per-call
  • Tool-level success/failure/latency
  • Distinct agent count — the vouch signal
Pillar 2Vouch-controlled
Synthetic probing
  • Vouch pings the endpoint every 60s
  • Protocol handshake verified
  • Schema conformance tested
  • Latency measured at p50 + p95
Vouch engine
Independent scoring
A
Vouched
Synthetic + community
B
Monitored
Synthetic + history
C
Conformant
Protocol valid
D
Reachable
HTTP 200
Developers
Select with confidence.
Search registry → inspect class → integrate
Agents
Route on live data.
REST · MCP · A2A — machine-readable
Endpoint owners
Earn certification.
Independent badge · Class A requires both pillars
Data sources — two independent pillarsVouch engine — scores and classifiesPublished — web · REST · MCP · A2A

Register the endpoint your agents depend on.

Submit any MCP, A2A, or REST endpoint. We probe it independently, score it honestly, and publish the result publicly. No login. No fee. No obligation to like what we find.

Register an endpointBrowse the registry →