# SpiderRating

> Independent security ratings for MCP servers, Claude skills, and AI tools.

SpiderRating is an independent security rating platform for the Model Context Protocol (MCP) ecosystem. We scan and score MCP servers, Claude skills, and AI tools using a three-layer scoring model called SpiderScore. As of March 2026, we have rated 15,923 MCP servers and skills.

## Core Concepts

- **SpiderScore**: A 0-10 composite security rating calculated as Description Quality (38%) + Security Analysis (34%) + Metadata Health (28%).
- **Grades**: A (9.0+), B (7.0-8.9), C (5.0-6.9), D (3.0-4.9), F (below 3.0).
- **Hard Constraints**: Critical security issues (reverse shell, credential exfiltration, command injection) force grade caps regardless of other scores.
- **Description Quality**: Evaluates 5 dimensions — Intent Clarity, Permission Scope, Side Effects, Capability Disclosure, Operational Boundaries. Maturity levels L1 (minimal) through L5 (production-grade).
- **Security Analysis**: Static analysis against 46+ rules covering command injection, path traversal, SQL injection, SSRF, prompt injection, data exfiltration, credential exposure, and more.
- **Metadata Health**: Provenance (40%), Maintenance (35%), Popularity (25%) — license clarity, last commit, GitHub stars/forks.

## Key Ecosystem Data (March 2026)

- Total rated: 15,923 MCP servers and skills
- Grade distribution: A: 0 (0.0%), B: 211 (1.3%), C: 11,045 (69.4%), D: 2,598 (16.3%), F: 2,069 (13.0%)
- Average SpiderScore: 5.26/10
- Average Description Quality: 4.76/10
- Average Security Score: 7.74/10
- Average Metadata Health: 4.92/10
- 64.0% of MCP servers lack sandbox configuration (the #1 ecosystem issue)
- No MCP server has yet achieved an A grade (9.0+)
- Only 1.3% of servers score B or higher

## Top Ecosystem Security Issues (March 2026)

1. sandbox.not_configured — 64.0% (10,197 servers)
2. token_leakage — 4.8% (757 servers)
3. ssrf — 2.2% (355 servers)
4. child_process_injection — 1.7% (270 servers)
5. no_input_validation — 1.6% (249 servers)
6. ts_path_traversal — 1.5% (245 servers)

## FAQ Pages (Topic-Organized)

- [All FAQs](https://spiderrating.com/faq): 18 common questions with schema.org FAQPage markup
- [MCP Security FAQ](https://spiderrating.com/faq/mcp-security): What MCP is, which servers are safe, common risks
- [Scoring FAQ](https://spiderrating.com/faq/scoring): How SpiderScore works, grades, rescans, disputes
- [API FAQ](https://spiderrating.com/faq/api): Decision API for AI agents
- [Runtime Protection FAQ](https://spiderrating.com/faq/runtime): Protecting live agents with hooks and proxies
- [Pricing FAQ](https://spiderrating.com/faq/pricing): Free, Pro ($49/mo), Business ($199/mo), Enterprise
- [Comparisons FAQ](https://spiderrating.com/faq/comparisons): vs SpiderShield, Glama, Smithery, mcp.so

## Pages

- [Methodology](https://spiderrating.com/methodology): How SpiderScore is calculated
- [Leaderboard — Top Rated](https://spiderrating.com/leaderboard/top): Highest-scoring MCP servers and skills
- [Leaderboard — Most Secure](https://spiderrating.com/leaderboard/security): Best security scores
- [Leaderboard — High Risk](https://spiderrating.com/leaderboard/risky): Servers with critical issues
- [Free Scanner](https://spiderrating.com/evaluate): Check any MCP server's security rating instantly
- [Directory](https://spiderrating.com/servers): Browse all 15,923 rated servers and skills
- [Blog](https://spiderrating.com/blog): Research reports and security analysis

## Blog Posts

- [97% of MCP Tools Don't Tell AI Agents When to Use Them](https://spiderrating.com/blog/state-of-mcp-documentation-2026-03) (2026-03-11)
- [OpenClaw 2026.3.1 Security Evaluation: Grade B](https://spiderrating.com/blog/openclaw-2026-3-1-security-evaluation) (2026-03-10)
- [We Scanned 200+ OpenClaw Skills. Here's What We Found.](https://spiderrating.com/blog/openclaw-skill-security-audit) (2026-03-10)
- [Introducing SpiderRating](https://spiderrating.com/blog/introducing-spiderrating) (2026-03-09)
- [How We Score MCP Servers: A Deep Dive](https://spiderrating.com/blog/how-we-score-mcp-servers) (2026-03-08)
- [The State of MCP Security in 2026](https://spiderrating.com/blog/state-of-mcp-security-2026) (2026-03-07)

## Decision API (For AI Agents — Free, No Auth)

**Use this when an AI agent needs to decide whether to install or invoke an MCP tool.**

`GET https://spiderrating.com/api/v1/decide/mcp-tool?slug={owner}/{repo}`

Returns a structured decision with verdict, confidence, recommendations, risk flags, and reasoning:

```json
{
  "slug": "samanhappy/mcphub",
  "decision": {
    "verdict": "NOT_RECOMMENDED",
    "confidence": 0.9,
    "summary": "Mcphub is not recommended — low score (2.0/10) with 7 critical issues."
  },
  "recommendations": [
    {"action": "run_in_container", "priority": "high", "reason": "..."}
  ],
  "not_recommended": ["running in production without container isolation"],
  "risk_flags": [
    {"type": "ts_sql_injection", "severity": "critical", "count": 7, "description": "..."}
  ],
  "reasoning": [
    {"factor": "overall_score", "impact": "negative", "weight": 0.5, "description": "..."}
  ],
  "provenance": {"data_sources": ["SpiderRating automated security scan"], "protocol_version": "v1.1"}
}
```

**Verdict enum**: `RECOMMENDED` | `CONSIDER` | `ALLOW_WITH_RISK` | `NOT_RECOMMENDED` | `AVOID` | `UNKNOWN`

## Other APIs (Free, No Auth Required)

- Check tool safety: `GET https://api.spiderrating.com/v1/public/check?tool={name}`
- Get server score: `GET https://api.spiderrating.com/v1/public/score/{owner}/{repo}`

## Data Files (JSON)

- [Ecosystem statistics](https://spiderrating.com/data/stats.json): Grade distribution, top issues, tool signal rates
- [All servers](https://spiderrating.com/data/servers.json): Full directory with scores and grades
- [Rankings](https://spiderrating.com/data/rankings.json): Sorted leaderboard data
- Individual server data: `https://spiderrating.com/data/servers/{owner}__{repo}.json`

## Open Source

SpiderRating is powered by [SpiderShield](https://pypi.org/project/spidershield/), an open-source MCP security scanner (MIT license).

```bash
pip install spidershield
spidershield scan https://github.com/owner/repo
```

## Contact

- Website: https://spiderrating.com
- API: https://api.spiderrating.com
- Security: security@spiderrating.com

Last updated: 2026-04-11 · Data as of March 2026 (2026-03-23)