MCP Trust Index

Public API

The whole index as JSON — free, unauthenticated, and versioned. Built for MCP gateways, agent IDEs and CI pipelines that need to check a server before it is installed.

Scores describe maintenance, not safety.Every response restates this in its meta block. If you surface these numbers to your own users, please carry the distinction through — and link to the methodology so they can see how a score was reached.

Base URL

https://mcp.klars.ai/api/v1

Endpoints

GET/api/v1/servers

The leaderboard, paginated and filterable.

Query parameters

  • limit — 1–100, default 50
  • offset — default 0
  • q — search name and description
  • risk — clean or flagged
  • verified — yes for servers with GitHub-verified activity data

GET/api/v1/server/{name}

One server, with its full per-signal score breakdown, every risk flag with its caveat, score history and registry record. URL-encode the name — it contains a slash.

/api/v1/server/ai.example%2Fmy-server

Example

curl -s "https://mcp.klars.ai/api/v1/servers?limit=5&risk=clean" \
  -H "user-agent: my-gateway/1.0 (contact@example.com)"

Response shape

{
  "data": [
    {
      "name": "ai.example/my-server",
      "score": 84.2,
      "rank": 12,
      "delta": 3.1,
      "unverifiedActivity": false,
      "riskFlagCount": 0,
      "highestSeverity": null
    }
  ],
  "pagination": { "total": 22014, "limit": 5, "offset": 0, "hasMore": true },
  "meta": {
    "scoreType": "activity-and-maintenance",
    "scoreRange": [0, 100],
    "documentation": "https://mcp.klars.ai/methodology",
    "licence": "CC-BY-4.0"
  }
}

Fair use

No key is required today. Responses are cached for five minutes and carry Cache-Control headers — please respect them rather than polling. Send a real User-Agent identifying your client so we can contact you rather than block you if something goes wrong. Clients that degrade the service for others may be rate-limited.

Authentication and metered tiers may be added later. The current endpoints stay versioned under /api/v1, so an integration written today will keep working.

Licence and attribution

Index data is licensed under CC BY 4.0. Use it commercially if you like — just credit “MCP Trust Index” with a link back to https://mcp.klars.ai.

Something missing?

If you need a field, a filter or a bulk export that is not here, email contact@klars.ai. We would rather add it than have you scrape the HTML.

← Back to the leaderboard