MCP 服务器

MEOK MCP Hardening MCP

Automated security red-team for any MCP server that scans manifests against OWASP LLM Top 10 and MCP-specific risks, returning a 0-100 hardening score and HMAC-signed report.

README

MEOK MCP Hardening MCP

Automated security red-team for ANY MCP server. Maps the OWASP LLM Top 10 (2025) plus 5 MCP-specific risks to a 0-100 hardening score and an HMAC-signed report.

🛡️ Part of the MEOK Governance Substrate (£499/mo) — combine with mcp-spec-compliance-mcp for spec-grade conformity AND security-grade hardening on every server you ship.

What it does

Every MCP server you publish ends up loaded inside someone's agent loop. That makes the manifest itself an attack surface. This MCP reads any server.json (or live MCP descriptor) and returns a structured security report covering:

Category	What we check
LLM01 Prompt injection	Instructional phrases in tool descriptions
LLM02 Insecure output	`eval` / `exec` / shell sinks
LLM05 Supply chain	Pinned versions, repo URL, provenance
LLM06 Secret disclosure	OpenAI / Anthropic / Stripe / GitHub / AWS / Slack keys in manifest
LLM07 Insecure plugin design	Missing name, over-broad tool surface
LLM08 Excessive agency	Destructive verbs (`delete`, `send`, `transfer`) without confirmation gate
LLM09 Overreliance	No license / homepage / metadata block
LLM10 Model theft	Public HTTP endpoint with no declared auth
MCP-S1 Tool-name spoofing	Non-ASCII characters / homoglyphs
MCP-S2 Roundtrip-input echoing	Untrusted-data sinks back to description
MCP-S3 Resource URI integrity	Plain `http://` resources
MCP-S4 Privilege exposure	`admin_` / `sudo_` tools on public surface
MCP-S5 Long-running tool gating	No cancel signal documented

Quick start

pip install meok-mcp-hardening-mcp
# or run with uvx (no install)
uvx meok-mcp-hardening-mcp

from server import audit
report = audit(your_server_json)
print(report.score(), report.grade())  # e.g. 87 "B"

Tools exposed

audit_server_json(server_json) — full report
audit_tool_description(tool) — one-tool deep scan
check_destructive_surface(server_json) — just LLM08 findings (CI gate)
check_supply_chain(server_json) — just LLM05 findings
list_owasp_findings() — rule map reference
generate_hardened_template() — passing-score starter manifest
sign_security_report(audit_result) — HMAC-seal for public verify

Scoring

Start at 100, subtract:

25 per critical finding
15 per high
8 per medium
3 per low

Grade: A ≥ 90 · B ≥ 80 · C ≥ 70 · D ≥ 60 · F otherwise.

Verify any signed report

Every signed report carries an HMAC tag. Verify at https://meok.ai/verify.

Why this exists

Every MCP author publishing to the Anthropic Registry, Smithery, Glama, or Awesome-MCP needs a clean security review. Every MCP consumer (Claude Desktop, Cursor, Windsurf) wants to verify what they're loading. This MCP is the seatbelt — free MIT, scriptable, signable.

Wire it up

// .mcp.json
{
  "mcpServers": {
    "meok-mcp-hardening": {
      "command": "uvx",
      "args": ["meok-mcp-hardening-mcp"]
    }
  }
}

Pricing

Self-host: free (MIT)
Starter: £29/mo — 1K hardening audits/month, signed report SLA
Pro: £79/mo — 10K audits/month, branded badge, public verify URL
Governance Substrate: £499/mo — bundled with 10 governance MCPs
A2A Substrate: £999/mo — bundled with all 12 A2A MCPs + attestation chain

Companion MCPs

mcp-spec-compliance-mcp — schema conformity audit
meok-mcp-cardgen-mcp — generate .well-known/mcp card
agent-prompt-injection-firewall-mcp — runtime injection defence
meok-aaif-agent-card-mcp — AAIF agent identity

💸 Try MEOK in 30 seconds — instant buy ladder

Tier	Price	What you get	Stripe
Smoke test	£1	Signed sample MCP-Hardening report + Article 50 PDF	https://buy.stripe.com/dRmcN75ScdQS7oh1Uc8k90U
Quick Kit	£9	EU AI Act Article 50 implementation guide (C2PA + EU-Icon)	https://buy.stripe.com/cNi00la8s1460ZT0Q88k90V
Founder Call	£29	30-min 1-on-1 with the founder	https://buy.stripe.com/8x228ta8s6oqbExaqI8k90W