free · no signup

Check your MCP server

Paste an MCP endpoint. In a few seconds you get an agent-readiness score and a prioritised list of fixes: broken schemas, undocumented parameters, missing safety hints, names some clients reject, and how many tokens your tool list costs every agent session.

Check

The ecosystem today

• 38.8% of servers have parameters with no description.
• 11.7% have tools that look state-changing but declare no destructiveHint.
• 0.8% use tool names that OpenAI-style function calling rejects.
• 68.1% declare no output schema on any tool.
• Tool-poisoning scan of every server: 0 with manipulative tool text, 9 that tell models to avoid alternatives, 28 that run arbitrary code or commands.

Full report: State of MCP agent-readiness & safety →

In CI

Fail a pull request when your server drops below a grade or picks up a tool-poisoning finding:

# .github/workflows/mcp-readiness.yml
on: [push, pull_request]
jobs:
  readiness:
    runs-on: ubuntu-latest
    steps:
      - run: pipx run glimind check https://your-server.example/mcp --min-grade B

Or locally: pip install glimind && glimind check <url> --fixes fixes.json to save corrected tool definitions.

Most agent-ready servers

1. Pipeworx Io Birdeye A 100 · 40 tools · ~13,703 tokens
2. 609nft Vibekit A 100 · 36 tools · ~4,079 tokens
3. Vibekit Bot Vibekit A 100 · 36 tools · ~4,079 tokens
4. Vlad Public Code Valem A 100 · 29 tools · ~8,106 tokens
5. Daniel Rafique Blackbook Advisor A 100 · 25 tools · ~5,768 tokens
6. Koynlabs Blackbook Advisor A 100 · 25 tools · ~5,768 tokens
7. Sqlguard A 100 · 24 tools · ~3,332 tokens
8. Google Cloud Container A 100 · 23 tools · ~5,522 tokens
9. Ktcod X402 Json Repair A 100 · 21 tools · ~5,303 tokens
10. Troyhunt Hibp A 100 · 17 tools · ~4,246 tokens

FAQ

What does the check do?

It connects to your MCP endpoint the way an agent would: one initialize handshake and one tools/list. It never calls a tool, sends no credentials and stores nothing about your traffic. Then it audits the advertised tools: schemas, descriptions, safety annotations, name compatibility, token cost, protocol version, and (for listed servers) 30-day uptime and breaking changes.

Why does token cost matter?

Many agents load every connected server's tool definitions into every request. A server whose tool list costs 20,000 tokens makes every call slower and more expensive, and crowds out the user's task. Across the ecosystem the median is ~2,308 tokens.

Why do tool annotations matter?

readOnlyHint and destructiveHint tell clients which tools are safe to run without asking and which change state. Without them a careful client asks the user about everything, and a careless one asks about nothing.

Is this free?

Yes, with no signup. Listed servers are re-audited automatically every few hours.

API: GET https://glimind.com/v1/readiness/<serverId> · POST https://glimind.com/v1/readiness/check {"url": "…"} · MCP tool audit_server. Hosts on our opt-out list are never contacted.