← MCP Doctor
State of MCP Quality 2026
mcp-doctor static scorecard plus live tool discovery against a catalog of 10 public MCP servers. Five servers connected at last run. This is an observation, not a certification.
v0 · observed 2026-07-10 · CLI 0.4.1 · commit 483af28 · 5/10 connected
| Rank | Server | Grade | Score | Tools | Tokens |
|---|---|---|---|---|---|
| 1 | MCP Filesystem | A | 100 | 14 | 1,997 |
| 2 | MCP Sequential Thinking | A | 100 | 1 | 996 |
| 3 | MCP Everything | A | 100 | 13 | 1,236 |
| 4 | MCP Memory | A | 90 | 9 | 1,040 |
| 5 | MCP Puppeteer | B | 83 | 7 | 612 |
Awards (v0)
- Best overall: MCP Filesystem (Grade A)
- Lowest token cost: MCP Puppeteer (612 tokens)
- Most tools: MCP Filesystem (14 tools)
Connection failures (v0)
- MCP Time: MCP error -32000: Connection closed
- MCP Fetch: MCP error -32000: Connection closed
- MCP Git: MCP error -32000: Connection closed
- MCP SQLite: MCP error -32000: Connection closed
- MCP Brave Search: MCP error -32000: Connection closed
Connection failures at last run were likely environment or network limits, not product defects.
v0 uses static agent-readiness checks. Task evals are optional and require your own model key.
Reproduce
npx --yes --package @coefficient-work/mcp-doctor@0.4.7 mcp-doctor benchmark -o ./reportsSource report on GitHub.