Index · litmus-v16
The Polygraph Index
Every grade we publish — MCP servers tested for behavior and ordered by adoption, Agent Skills scanned for safety. What each one does, not what its README claims.
106 MCP servers graded, ranked by adoption · 13 live endpoints (hosted, egress unverified) · 3 skills scanned · adoption data as of 2026-07-11. A grade is a measurement, not a guarantee; you can re-run the open harness yourself.
| # | Server | Grade | Checks | Adoption |
|---|---|---|---|---|
| 1 | npm/@playwright/mcp | A | 01✓02✓03✓04✓ | 100/10024.8M npm/mo |
| 2 | npm/@modelcontextprotocol/server-filesystem | A | 01✓02✓03✓04✓ | 87/1001.72M npm/mo |
| 3 | npm/@upstash/context7-mcp | A | 01✓02✓03✓04✓ | 84/1004.31M npm/mo |
| 4 | npm/@modelcontextprotocol/server-sequential-thinking | A | 01✓02✓03✓04✓ | 82/100573K npm/mo |
| 5 | npm/@modelcontextprotocol/server-memory | A | 01✓02✓03✓04✓ | 79/100440K npm/mo |
| 6 | npm/firecrawl-mcp | A | 01✓02✓03✓04✓ | 79/100545K npm/mo |
| 7 | npm/n8n-mcp | A | 01✓02✓03✓04✓ | 78/100547K npm/mo |
| 8 | npm/@modelcontextprotocol/server-github | A | 01✓02✓03✓04✓ | 78/100616K npm/mo |
| 9 | npm/@notionhq/notion-mcp-server | A | 01✓02✓03✓04✓ | 77/100587K npm/mo |
| 10 | npm/@antv/mcp-server-chart | A | 01✓02✓03✓04✓ | 72/10090.8K npm/mo |
| 11 | npm/@modelcontextprotocol/server-puppeteer | A | 01✓02✓03✓04✓ | 72/100122K npm/mo |
| 12 | npm/exa-mcp-server | A | 01✓02✓03✓04✓ | 71/100103K npm/mo |
| 13 | npm/@browserbasehq/mcp | A | 01✓02✓03✓04✓ | 70/10063.5K npm/mo |
| 14 | npm/tavily-mcp | A | 01✓02✓03✓04✓ | 68/100132K npm/mo |
| 15 | npm/@21st-dev/magic | A | 01✓02✓03✓04✓ | 67/10069.8K npm/mo |
| 16 | npm/@negokaz/excel-mcp-server | A | 01✓02✓03✓04✓ | 67/10061.8K npm/mo |
| 17 | npm/@netlify/mcp | A | 01✓02✓03✓04✓ | 66/10049.8K npm/mo |
| 18 | npm/mcp-server-kubernetes | A | 01✓02✓03✓04✓ | 66/10081.5K npm/mo |
| 19 | npm/@alchemy/mcp-server | A | 01✓02✓03✓04✓ | 62/10035.1K npm/mo |
| 20 | npm/@odgrim/mcp-datetime | A | 01✓02✓03✓04✓ | 62/10048.6K npm/mo |
✓ pass✕ fail– not runC-01 tool-output injection · C-02 egress overreach · C-03 sensitive-data handling · C-04 adversarial-input handling
Ranked by the adoption score (0–100, shown at right above monthly downloads) — a composite of downloads (npm / PyPI), GitHub stars, dependents and release velocity. It measures reach, not safety: the litmus grade is the only safety verdict. Grades come from the open litmus harness; you can request a grade for a server, or read the methodology.
| Skill | Grade | Checks |
|---|---|---|
| productclank-campaigns | A | S01✓S03✓S04✓ |
| productclank-agent-participation | A | S01✓S03✓S04✓ |
| skills/pdf | A | S01✓S03✓S04✓ |
✓ pass✕ fail– not runS-01 prompt-injection · S-03 data-exfiltration instructions · S-04 dangerous bundled commands
Skills are graded by a static safety scan (litmus-skill-v2) of the SKILL.md and its bundled files — A/B/D/F, no adoption ranking. It reads the skill’s text, not its runtime behavior, so it’s a static read rather than behavioral proof. Read the methodology.