MCP Prompt Injection Scanner

Detects prompt injection attacks in MCP tool inputs — OWASP LLM Top 10 coverage, real-time scanning, severity scoring for AI agent security

meok-mcp-injection-scan-mcp MCP server MCP Registry PyPI

meok-mcp-injection-scan-mcp MCP server

meok-mcp-injection-scan-mcp

PyPI version PyPI downloads License: MIT MEOK AI Labs

Scan any MCP server for the prompt-injection / tool-poisoning / SSRF class disclosed in the April 2026 CVE wave.

pip install meok-mcp-injection-scan-mcp

Why this exists

April 2026 was a bad month for MCP. Anthropic published a "by-design" MCP RCE class affecting ~7,000 public servers (~150M downloads). mcp-server-git shipped a CVE chain. DockerDash got popped by an injection chain. Tool-description prompt injection ("tool poisoning") was demonstrated against every major MCP host.

If you run an MCP server in production, or you're auditing one before adoption, you need a fast scan that flags the patterns the April 2026 disclosures target. This MCP is that scan.

What it checks

30+ canonical rules across 5 severity tiers:

  • CRITICAL — direct RCE, system-prompt override, credential exfil patterns, shell metachars in defaults, file:// / internal-network URLs (the DockerDash 169.254.169.254 metadata-pivot vector).
  • HIGH — encoded payloads, imperative directives at the agent, supply-chain prompts, env-var references, tool shadowing.
  • MEDIUM — urgency / authority language, additionalProperties=true, unbounded strings, tool-name impersonation.
  • LOW — over-long descriptions, zero-width / bidi-override chars (the U+202E PoC vector).

Coverage maps to: OWASP LLM Top 10, GenAI Red Team v1, the April 2026 Anthropic MCP RCE disclosure, and the mcp-server-git CVE chain.

Tools exposed

ToolPurpose
scan_mcp_url(url)Fetch a remote MCP server's tool listing and scan it
audit_tool_descriptions(tools_json)Scan a pasted JSON tool list (auth-walled servers)
signed_safety_report(subject, findings_json, score, note)Issue a procurement-grade signed cert (Pro tier)
list_rules()Inspect the full rule catalogue before subscribing
pricing()Subscribe links + tier comparison

Pricing

TierPriceWhat you get
Free£05 scans / day, no signed reports
Starter£29/moUnlimited scans + signed reports
Pro£79/mo+ scheduled rescans + 48h support
Enterprise£1,499/mo+ custom rule packs + 4h SLA

Every signed cert lives at https://meok-attestation-api.vercel.app/verify/<cert_id> — auditors and procurement teams confirm without an account.

What you do NOT get

This is a static-pattern scanner. It does not run dynamic taint analysis, fuzz the server with adversarial inputs, or replace a human red-team. It is the first 80% of the audit, in 5 seconds, for free.

Built by MEOK AI Labs

Solo founder. London. 234 MCP packages on PyPI. Live signing infrastructure at meok-attestation-api.vercel.app. Storefront councilof.ai. Get the catalogue: https://meok-attestation-api.vercel.app/catalogue.


Distribution channels

Máy chủ liên quan

NotebookLM Web Importer

Nhập trang web và video YouTube vào NotebookLM chỉ với một cú nhấp. Được tin dùng bởi hơn 200.000 người dùng.

Cài đặt tiện ích Chrome