How we rank

Rankings are computed within each category and warehouse so a 150k-star general agent doesn't bury a focused Snowflake MCP server.

DataMCP score (0–100)

• Popularity (0–60): 12 × log₁₀(GitHub stars + 1), capped — logarithmic so stars matter but don't dominate.
• Freshness (0–30): 30 minus 1 point per 6 days since last commit on the default branch.
• Official (0–10): maintained by the vendor of the target system.
• Archived: −40.

Warehouse pages

Warehouse rankings add a specialist bonus of 24 ÷ (number of targets covered), so a dedicated Snowflake MCP server outranks a framework that supports everything.

Data refresh

A GitHub Action runs scripts/refresh.py nightly, updating stars, last commit, license and archived status, then rebuilds this static site.

Testing

Planned: each MCP server is launched in CI against DuckDB/Postgres fixtures and checked for list_tools, a sample read query, and whether writes are possible (to verify the read-only/write label). Results appear as a "Tested" badge with date.

Inclusion

Only real, public projects with a verifiable repo. Paid placement is always labeled and never changes score.