Full comparison table

All scores are on a 0–100 scale. The average across 910 database tools is 29.6. The tools below represent the top tier — all scoring above 70.

Top 10 breakdown

#1 — motherduckdb/mcp-server-motherduck (89.06)

motherduckdb/mcp-server-motherduck leads the database category with a score of 89.06. DuckDB runs in-process — no separate server, no network round-trips. That makes it the fastest option for analytical queries over local files (Parquet, CSV, JSON) or in-memory datasets. MotherDuck support extends it to cloud scale when needed. 80% issue close rate signals an active maintainer. Best choice if your agent is doing data analysis or ETL work.

#2 — redis/mcp-redis (88.84)

redis/mcp-redis is the official Redis Labs server scoring 88.84. It exposes Redis data structures — strings, hashes, sets, sorted sets, streams, and search — through natural language. 68% issue close rate and official vendor backing means this will track breaking protocol changes. The right choice for any agent working with caching, session state, rate limiting, or pub/sub messaging.

#3 — mongodb-js/mongodb-mcp-server (88.56)

mongodb-js/mongodb-mcp-server is MongoDB's official server at 88.56. It's the most-starred tool in the top 3 (965 stars) and has the highest issue close rate of the three at 84% — meaning reported problems actually get resolved. Covers local MongoDB instances and cloud-hosted Atlas clusters. If your agent stack includes document storage, this is the reference implementation to use.

#4 — benborla/mcp-server-mysql (85.22)

benborla/mcp-server-mysql is the most-starred MySQL MCP server with 1,361 stars and a score of 85.22. It's read-only by design, which is a deliberate safety constraint for production use — your agent can inspect schemas and run SELECT queries but can't accidentally mutate data. If you need write access, fork it or pair it with a write-capable layer.

#5 — neondatabase/mcp-server-neon (78.58)

neondatabase/mcp-server-neon is Neon's official server scoring 78.58. It covers both the Neon management API (create/delete databases, manage branches) and raw SQL query execution — two distinct use cases in one server. 564 stars and TypeScript implementation. Best for teams running serverless or edge Postgres on Neon's platform.

#6 — pgEdge/pgedge-postgres-mcp (76.98)

pgEdge/pgedge-postgres-mcp scores 76.98 with a Go implementation. It includes a natural language agent CLI in addition to the standard MCP interface — useful if you want to test queries locally before wiring them into an agent workflow. 62% issue close rate. Lower star count (90) but actively maintained by the pgEdge team.

#7 — dbt-labs/dbt-mcp (76.04)

dbt-labs/dbt-mcp is the official dbt Labs server at 76.04. It sits between your dbt project and any agent — exposing models, tests, and query execution through MCP. 507 stars and 68% issue close rate. The only data engineering tool in this list. Essential if your stack uses dbt for transformation; skip it if it doesn't.

#8 — HenkDz/selfhosted-supabase-mcp (75.44)

HenkDz/selfhosted-supabase-mcp scores 75.44 with a 100% issue close rate — every reported issue resolved. Built specifically for teams running Supabase on their own infrastructure (not Supabase Cloud). TypeScript, 118 stars. A niche tool, but the right one if self-hosted Supabase is your stack.

#9 — runekaagaard/mcp-alchemy (74.93)

runekaagaard/mcp-alchemy is the broadest database coverage tool in this list at 74.93. It uses SQLAlchemy under the hood — which means it works with any SQLAlchemy-compatible database: Postgres, MySQL, SQLite, MSSQL, Oracle, and more. 100% issue close rate. 397 stars. Best for polyglot database environments.

#10 — bytebase/dbhub (70.71)

bytebase/dbhub is the most-starred database MCP server in the entire index with 2,335 stars, though its composite score (70.71) trails the top tier. Covers Postgres, MySQL, SQL Server, MariaDB, and SQLite from one zero-dependency server. 95% issue close rate. Strong pick if your infrastructure runs multiple database engines and you want one server to rule them.

Choosing by database type

Scoring methodology

AgentRank scores combine five GitHub signals: stars (15%), commit freshness (25%), issue close rate (25%), contributor count (10%), and inbound dependents (25%). Scores are updated nightly from live crawl data. The average score across all 910 database-category tools is 29.6. Full methodology →