Scan · Score · Install — Your MCP Server Toolkit
The missing security and quality toolkit for the Model Context Protocol ecosystem. Scan MCP servers for vulnerabilities, score their quality, compare alternatives, and install with confidence.
| 📦 Registry | 🔒 Security Checks | 📊 Quality Dimensions |
|---|---|---|
| 100+ servers scored | 10 detection engines | 5 scoring categories |
The MCP ecosystem is exploding — hundreds of servers, but no way to know which ones are safe, well-maintained, or even work. MCP Doctor is your one-stop CLI to navigate the MCP jungle.
Before installing an MCP server, run it through the Doctor.
In the first 6 months of 2026, security researchers filed 30+ CVEs targeting MCP servers, clients, and infrastructure:
- CVE-2026-44717 —
eval()RCE in mcp_calculate_server (CVSS 9.8) - CVE-2026-49257 — Default no-auth + 0.0.0.0 binding in mcp-pinot (CVSS 10.0)
- CVE-2026-46519 — Access control bypass in mcp-server-kubernetes (presentation-layer only filtering)
- CVE-2026-32814 — MCP SDK authentication bypass, 3 Indian AI startups compromised
- CVE-2026-25536 — Concurrent session data leakage in mcp-handler (Vercel)
A study of 39,884 open-source MCP servers found 106 zero-day vulnerabilities. OX Security found that 82% of MCP implementations have path traversal vulnerabilities.
MCP Doctor catches these before you deploy them.
| Feature | MCP Doctor | Cisco mcp-scanner | Snyk Agent Scan | Oxvault | mcp-shield | mcpsec |
|---|---|---|---|---|---|---|
| Security scanning | ✅ 10 engines | ✅ YARA+LLM+API | ✅ LLM+guardrails | ✅ SAST+runtime | ✅ 20 rules | ✅ fuzzer |
| Quality scoring (0-100) | ✅ | ❌ | ❌ | ❌ | ✅ (A-F) | ❌ |
| Curated registry (100+) | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ |
| Server comparison | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ |
| One-command install | ✅ | ❌ | ❌ | ❌ | ❌ | ❌ |
| CI/CD integration | ✅ | ✅ | ✅ | ✅ (SARIF) | ❌ | ❌ |
| Runtime proxy mode | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
| REST API server | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
| Local / no API dependency | ✅ | ✅ | ❌ (needs token) | ✅ | ✅ | ✅ |
| SARIF output | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ |
How MCP Doctor fits: Cisco mcp-scanner and Snyk Agent Scan are excellent for auditing existing deployments. Oxvault is best for deep SAST of MCP source code. MCP Doctor is for the pre-install workflow — discover, compare, and install servers with confidence. Most teams need both: a scanner for what's deployed and a discovery tool for what to deploy next.
- 🔒 Security Scan — 8 detection engines: prompt injection, path traversal, credential leakage, network exfiltration, command injection, SSRF, supply chain, and excessive permissions
- 📊 Quality Score — Automated 0-100 scoring based on: security, maintenance, documentation, testing, community
- ⚖️ Compare — Side-by-side comparison of two MCP servers across all quality dimensions
- 📦 Registry — Curated database of 100+ popular MCP servers with pre-computed scores and categories
- ⚡ Quick Install —
mcp-doctor install <server>— one command to scan + score + install - 🩺 CI/CD Audit —
mcp-doctor audit <server>— full report with exit codes for pipelines - 📈 Stats — Registry overview with category breakdowns and averages
# Install
pip install mcpdoctor-cli
# Or install directly from GitHub (if PyPI not yet available)
pip install git+https://github.com/xlyoung/mcp-doctor
# Scan an MCP server for security issues
mcp-doctor scan @modelcontextprotocol/server-filesystem
# Score a server's quality (0-100)
mcp-doctor score modelcontextprotocol/servers
# Compare two servers side-by-side
mcp-doctor compare @modelcontextprotocol/server-postgres mcp-server-mysql
# See top-rated servers by category
mcp-doctor list --category database --sort score
# Install with confidence (blocks on critical security issues)
mcp-doctor install @modelcontextprotocol/server-github
# Full audit for CI/CD (exits non-zero on failure)
mcp-doctor audit @modelcontextprotocol/server-filesystem --threshold 60
mcp-doctor audit @modelcontextprotocol/server-filesystem --json # machine-readable
# View registry stats
mcp-doctor stats$ mcp-doctor scan @modelcontextprotocol/server-filesystem
🩺 MCP Doctor — Security Scan
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Target: @modelcontextprotocol/server-filesystem
🔒 Security Issues Found: 3
├── 🟠 MEDIUM Unconstrained Parameters
│ └── 6 string params lack maxLength/pattern constraints
│ → path (read_file, write_file, edit_file, ...)
├── 🟡 LOW Excessive Permissions
│ └── Configurable access control but no default restrictions
└── 🟡 LOW No Input Sanitization
└── Path parameters accept ../ traversal sequences
⚠️ Security Score: 72/100 (Fair)
Use --json for machine-readable output
Use --threshold 80 to enforce minimum score in CI/CD
Each MCP server receives a score from 0-100:
| Category | Weight | What it checks |
|---|---|---|
| Security | 35% | Prompt injection, path traversal, network safety |
| Maintenance | 25% | Commit frequency, issue response time, release cadence |
| Documentation | 15% | README quality, API docs, examples |
| Testing | 15% | Test coverage, CI status, integration tests |
| Community | 10% | Stars, contributors, ecosystem integrations |
MCP Doctor scans for common MCP vulnerability patterns:
- Prompt Injection — User input flowing directly into LLM prompts without sanitization
- Path Traversal — File system tools accepting
../in paths without validation - Credential Leakage — API keys, tokens, or secrets exposed in logs or responses
- Network Exfiltration — Suspicious external endpoints (webhook.site, ngrok tunnels)
- Supply Chain — Typosquatting, abandoned dependencies, known CVEs
- SSRF — Server-side request forgery via unvalidated URL parameters
- Command Injection — Shell command execution with unsanitized input
- Excessive Permissions — Auth bypass flags, unrestricted filesystem globs, world-readable permissions
- Unconstrained Parameters — String params without maxLength/pattern/enum (DoS + injection chain risk)
- Tool Access Control Bypass — Presentation-layer filtering without execution-layer enforcement (CVE-2026-46519 pattern)
Can't decide between two MCP servers? Compare them:
$ mcp-doctor compare @modelcontextprotocol/server-postgres mcp-server-mysql
MCP Server Comparison
┏━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Metric ┃ postgres ┃ mysql ┃
┡━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Overall Score │ 90/100 │ 78/100 │
│ Security │ 90 │ 80 │
│ Maintenance │ 95 │ 60 │
│ Documentation │ 70 │ 50 │
│ Testing │ 55 │ 50 │
│ Community │ 95 │ 80 │
│ Security Issues │ 0 │ 0 │
└──────────────────┴───────────┴───────────┘
✔ Recommendation: @modelcontextprotocol/server-postgres (score 90 vs 78)Browse the full registry: REGISTRY.md
Categories:
- 🗄️ Database — PostgreSQL, MySQL, SQLite, Redis, MongoDB, Qdrant, ClickHouse, Supabase, Neon, Google Toolbox
- 📁 File System — Local files, S3
- 🌐 Web — Brave Search, Playwright, Puppeteer, Firecrawl, Chrome MCP, BrowserMCP, bb-browser
- 🔧 DevOps — GitHub (official & MCP), GitLab, Docker, Kubernetes, Git, XcodeBuild
- 💬 Communication — Slack, Discord, Gmail, Telegram, WhatsApp
- 🧠 Knowledge — Notion, Obsidian, Memory, Confluence
- 🤖 AI/ML — FastMCP, Hugging Face, Replicate, Ollama
- 🔒 Security — Semgrep, NVD CVE, GhidraMCP, MCP Audit, MCPSec, HexStrike AI
- 🎨 Design — Figma Context MCP
- 📋 Productivity — Google Maps, Linear, Airtable, Google Calendar, Todoist, Kreuzberg
- 🖥️ System — DesktopCommanderMCP, Windows-MCP
- 🛠️ Developer Tools — MCP Inspector, git-mcp, MCP Registry
- ☁️ Cloud — AWS (awslabs), ActivePieces
- 🏗️ Framework — mcp-use
- 🔄 Automation — ActivePieces
git clone https://github.com/xlyoung/mcp-doctor.git
cd mcp-doctor
pip install -e ".[dev]"
pytestFound an MCP server that should be in the registry? Open an issue or submit a PR!
See CONTRIBUTING.md for guidelines.
Found a security vulnerability? See SECURITY.md for responsible disclosure guidelines.
See CHANGELOG.md for release history.
MIT — do whatever you want with it.
Built for the MCP community. Star ⭐ if you find it useful!