mcpscan MCP server

MCPScan

Scan your MCP servers in 30 seconds and collect production-readiness evidence.

MCPScan is a free CLI security scanner and audit-readiness toolkit for Model Context Protocol servers. Point it at a Claude Desktop style MCP config, a local server command, or a remote HTTPS/SSE endpoint and get an A-F grade with actionable findings, remediation notes, and CI-friendly evidence.

Launch / First Revenue

The first-revenue motion is a fixed-scope MCP readiness audit, supported by the CLI, public sample report, intake pages, and local launch consoles.

Start here from the repo:

npm run launch:next

That command opens the current founder action stack, GitHub billing, GitHub Actions, Spaceship domain search, and Stripe Payment Links. It prints live launch status and shows the next action ladder. It does not buy, publish, send, apply links, or create customer files.

Key surfaces:

• Founder approval packet
• Launch cockpit
• Final founder click path
• GitHub issue action board
• Public landing page
• Sample report
• Secure intake guidance

Current launch verification:

npm run launch:next
npm run launch:verify

Expected pre-approval warnings are GitHub Actions billing or account lock, checkout links, custom domain, and custom security email until the founder-only clicks are complete.

npx mcpscan scan ./claude_desktop_config.json
npx mcpscan scan --server "npx -y @modelcontextprotocol/server-filesystem /tmp"
npx mcpscan scan --url https://mcp.example.com/sse
npx mcpscan report ./scan-results.json --format html --output report.html
npx mcpscan scan ./mcp-config.json --format sarif --output mcpscan.sarif

What It Checks Today

This MVP ships the CLI first:

• AUTH-001: no authentication hints detected
• AUTH-002: bearer/API token validation cannot be confirmed
• AUTH-003: API key exposure in URL, command, or static config
• AUTH-004: static session or token reuse indicators
• INPUT-001: command-injection prone tool arguments
• INPUT-002: SQL-injection prone tool arguments
• INPUT-003: path-traversal prone file arguments
• INPUT-004: URL-fetching tool SSRF risk
• INPUT-005: weak argument-shape constraints
• OUTPUT-001: secret and sensitive data patterns
• OUTPUT-002: excessive data exposure
• OUTPUT-003: error disclosure and stack-trace indicators
• TRANSPORT-001: TLS probe for remote HTTPS targets
• TRANSPORT-002: stdio process-isolation heuristics
• TRANSPORT-003: SSE/WebSocket origin validation heuristics
• TOOL-001: tool description prompt-injection language
• TOOL-002: excessive permission language
• TOOL-003: shadow-tool impersonation language
• TOOL-004: tool-name collision detection
• CONFIG-001: overly permissive CORS for URL targets
• CONFIG-002: debug-mode indicators
• CONFIG-003: default or example credentials

Each finding includes evidence, remediation, and MCP security references.

CLI

mcpscan scan <config-or-server>
  --checks auth,input,output
  --severity critical,high
  --format json|table|html|markdown|sarif
  --output ./report.json
  --ci
  --min-grade B
  --timeout 30
  --verbose
  --no-color

CI

name: MCP Security Scan
on: [push, pull_request]
jobs:
  scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: Davidleeops/mcpscan@main
        with:
          target: ./mcp-config.json
          min-grade: B

See docs/GITHUB_ACTION.md for SARIF/code-scanning upload examples.

Landing Page

A minimal static commercial landing page is available in landing/. See landing/README.md for local preview and deployment options, and docs/LANDING_PAGE.md for the pre-publish checklist.

Why Now

MCP adoption is accelerating, but many servers expose powerful tools through weak auth, broad filesystem/network access, and unsafe prompt-facing tool metadata. MCPScan is built to become the independent, developer-friendly path from local MCP scans to production-readiness audits, OWASP MCP Top 10 coverage, NSA advisory alignment, and buyer-facing governance evidence.

Comparison

Tool	A-F grade	Free local CLI	Dashboard-ready reports	CI threshold
MCPScan	Yes	Yes	JSON/HTML/Markdown/SARIF now, SaaS later	Yes
Snyk Agent Scan	Enterprise-focused	Limited	Yes	Yes
Cisco mcp-scanner	No	Yes	No	Basic
Promptfoo	Eval-focused	Yes	Eval reports	Partial

Development

npm install
npm run build
npm test
node packages/cli/dist/index.js scan packages/cli/test/fixtures/vulnerable-config.json

The CLI does not send scan results anywhere. The future dashboard should only receive reports through an explicit upload/API action.

Use MCPScan only on systems, configurations, repositories, and MCP endpoints you own or are authorized to review.