grahamrowe82/mcp-quality-index
Daily-updated quality scores for 12,653+ MCP servers. Machine-readable JSON. No auth required.
Platform-specific configuration:
{
"mcpServers": {
"mcp-quality-index": {
"command": "npx",
"args": [
"-y",
"mcp-quality-index"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
[](https://github.com/grahamrowe82/mcp-quality-index/commits/main) [](#datasets) [](LICENSE)
Daily-updated quality scores for 12,653+ MCP server repositories. Machine-readable JSON. No auth required.
Every MCP registry today is a flat catalog — name, description, maybe a link. None of them tell you whether a server is maintained, adopted, or safe to depend on.
This dataset scores every MCP-domain repository on GitHub across four dimensions:
| Component | Range | What it measures | |-----------|-------|-----------------| | Maintenance | 0-25 | Commit activity, push recency, archived status | | Adoption | 0-25 | Stars, package downloads, reverse dependents | | Maturity | 0-25 | License, published package (PyPI/npm), repo age | | Community | 0-25 | Forks, fork-to-star ratio |
The composite quality score (0-100) classifies each server into a tier:
| Tier | Score | Count | Description | |------|-------|-------|-------------| | Verified | 70+ | 53 | Strong across all dimensions | | Established | 50-69 | 680 | Solid but gaps in one area | | Emerging | 30-49 | 3,078 | Early-stage, some traction | | Experimental | <30 | 8,842 | Minimal signals |
Risk flags are computed per repo: archived, no_license, stale_6m, no_package, no_dependents.
See METHODOLOGY.md for the full scoring model, and `schema/` for field-level documentation.
All files are in `data/`:
| File | Records | Description | |------|---------|-------------| | `mcp-scores.json` | 12,653 | Quality scores with component breakdown and risk flags | | `mcp-repos.json` | 12
Loading reviews...