Introducing SpiderRating: Independent Security Ratings for MCP Servers
Why MCP servers need security ratings
The Model Context Protocol (MCP) ecosystem is growing fast. Thousands of servers now expose tools to AI agents — from database access to file system operations, from API integrations to code execution. But how do you know which servers are safe to use?
Unlike npm packages or Docker images, MCP servers don't have a standardized quality signal. There's no "npm audit" for MCP tools. SpiderRating changes that.
How SpiderRating works
Every MCP server is scored across three dimensions:
- Description Quality (35%) — Are tool descriptions clear, unambiguous, and complete? Do they disclose side effects and permissions? LLMs rely on these descriptions to decide when and how to use a tool.
- Security Analysis (35%) — Static analysis against 46 security rules covering malicious patterns (reverse shells, credential theft, prompt injection), suspicious behaviors (typosquatting, toxic data flows), and configuration issues (missing auth, weak sandboxing).
- Metadata Health (30%) — Provenance signals like license, maintenance activity, popularity, and repository quality.
Grades and hard constraints
Scores map to letter grades: A (8.0+), B (6.0+), C (4.0+), D (2.0+), F (below 2.0).
Critical security issues trigger hard constraints that cap the maximum grade — a server with a detected reverse shell can never score above F, regardless of how good its descriptions are.
Open source and reproducible
SpiderRating is powered by TeeShield, our open-source MCP security linter. Every scan is deterministic: same input, same output. No AI-in-the-loop scoring, no subjective judgments. You can run the same analysis locally:
pip install teeshield
teeshield scan https://github.com/owner/repo
What's next
We're actively scanning the MCP ecosystem and accepting community submissions. If you maintain an MCP server, submit it for evaluation and see how it scores.
We believe transparency drives better security practices. When maintainers know their server will be publicly rated, they're more likely to invest in quality descriptions and secure implementations.