Database MCP Servers — Postgres, MySQL, SQLite, MongoDB (2026) | Skills and Servers
Subscribe
Category · 12 servers

Database MCP Servers — Postgres, MySQL, SQLite, MongoDB, Redis (2026)

Every MCP server in our directory that connects an AI assistant to a database — relational (Postgres, MySQL, SQL Server), embedded (SQLite, DuckDB), NoSQL (MongoDB, Cassandra), key-value (Redis, DynamoDB), warehouses (Snowflake, BigQuery), and search (Elasticsearch). Each server has been installed, tested against a real database, and documented with auth setup notes for Claude Desktop, Claude Code, Cursor, and Cline.

12
Servers
5
Sub-types
4
Clients
100%
Human-tested
SS Curated by Skills and Servers Editorial Last updated August 17, 2026

1What this category covers

Database MCP servers expose a database to an AI assistant through the Model Context Protocol. They translate the assistant's natural-language requests into safe database operations (schema introspection, SELECT queries, sometimes writes) and return the results back to the assistant. The AI never talks to the database directly; the MCP server is the middleware that enforces auth, connection pooling, query limits, and safety guards.

The value the MCP layer adds is safe abstraction. Plain LLM output that connects to a database can hallucinate tables, run unbounded queries, or execute writes when only reads were intended. A well-designed database MCP server enforces boundaries — read-only by default, explicit schemas, connection limits, statement timeouts, query complexity caps. For real databases with real data, this discipline is essential.

What lives here vs. elsewhere: if the server's primary job is talking to a database engine, it belongs here. If it's about analytics on top of a database (BI tools, dashboards), see Data & Analytics. If it's about file-based data (CSVs, Parquet in a bucket), see Files & Storage. Servers with dual-purpose (Snowflake is both warehouse and analytics platform) appear in both hubs.

2When you should reach for a Database MCP server

  • You want to query production or staging databases in natural language without hand-writing SQL every time.
  • You're building AI workflows that read from and (sometimes) write to a database as part of a larger flow.
  • You need schema-aware code generation — the AI reads the real schema before generating queries or ORM code.
  • You want to give an AI assistant safe, read-only access to a database for support or investigation workflows.
  • You're prototyping an application against a real data source and want AI to explore the schema and data with you.

3Featured servers this month

Our editorial picks — heaviest tested, highest reader ratings.

Editor's pick

Postgres MCP Server

The reference implementation for relational databases. Best-tested. Handles connection pooling, schema introspection, safe read-only mode, and query complexity limits out of the box.

SQL·Tested Aug 15·Ahmed R.
Most requested

MySQL MCP Server

MySQL / MariaDB / Aurora MySQL counterpart. Same tool surface as Postgres MCP. Handles connection pooling, prepared statements, and read-replica routing.

SQL·Tested Aug 13·Sana K.
Highest reliability

SQLite MCP Server

Local file-based database. Great for prototyping data models before productionizing, working with SQLite-based apps (browsers, mobile apps), or as a lightweight cache for AI workflows.

SQL·Tested Aug 12·Ahmed R.

4All 12 Database servers

Filter by sub-type, search by name, sort by recency or alphabetical.

Showing 12 of 12 servers
SQLAug 15

Postgres MCP Server

PostgreSQL 12–17. Connection pooling, schema introspection, read-only mode, statement timeouts.

ARAhmed R.View →
SQLAug 13

MySQL MCP Server

MySQL 8+, MariaDB, Aurora MySQL. Prepared statements, read-replica routing, character-set aware.

SKSana K.View →
SQLAug 12

SQLite MCP Server

File-based database. Perfect for prototyping, mobile-app inspection, and lightweight AI workflow caches.

ARAhmed R.View →
SQLAug 10

SQL Server MCP Server

Microsoft SQL Server 2019+, Azure SQL. T-SQL dialect, Windows auth support, always-encrypted aware.

SKSana K.View →
NoSQLAug 14

MongoDB MCP Server

MongoDB 6+, Atlas. Aggregation pipeline support, index inspection, safe read-only mode by default.

ARAhmed R.View →
NoSQLAug 8

Cassandra MCP Server

Apache Cassandra 4+, DataStax. CQL query support, partition-aware queries, replication-factor aware.

SKSana K.View →
NoSQLAug 6

DynamoDB MCP Server

AWS DynamoDB. IAM auth, GSI-aware queries, cost-estimate before running expensive scans.

ARAhmed R.View →
WarehouseAug 11

Snowflake MCP Server

Snowflake. Warehouse selection, credit estimation, VPS-safe read-only mode. Cross-database queries.

SKSana K.View →
WarehouseAug 9

BigQuery MCP Server

Google BigQuery. Cost-estimate before queries, dry-run mode, INFORMATION_SCHEMA introspection.

ARAhmed R.View →
Key-ValueAug 7

Redis MCP Server

Redis 7+. Key patterns, TTL inspection, Cluster-aware, pub/sub-safe operations only by default.

SKSana K.View →

5Related Skills

AI Skills our team pairs with Database MCP servers for full workflows.

6Related Learn guides

Longer-form tutorials for the concepts behind these servers.

7Comparisons worth reading

Head-to-head reviews that touch on Database workflows.

8Frequently asked questions

Yes with the right setup. Best practice: create a dedicated database user with read-only permissions on only the schemas the AI needs access to; set explicit statement timeouts and connection limits; enable query logging so you have an audit trail. All of the SQL MCP servers in this category default to read-only mode unless explicitly configured otherwise. See our safe-setup guide for the full recipe.
Not with the recommended read-only setup. The MCP server enforces read-only mode at the connection level, and the database user should not have DROP/DELETE/UPDATE permissions to begin with (defense in depth). For workflows that genuinely need writes, use a separate MCP server instance with a write-enabled user and explicit human approval on destructive operations. Never share credentials between read-only and write-enabled contexts.
For SQL specifically, Claude Opus 4.7 and GPT-5 are near-tied at the top — see our detailed comparison. Claude is slightly better on complex CTEs and multi-database schemas; GPT-5 is slightly better on window functions and PIVOT patterns. For general database exploration and schema-aware code generation, either works well. For high-volume automation, DeepSeek V4 is competitive at much lower cost.
Yes — all servers in this category work with Claude Desktop, Claude Code, Cursor, and Cline as of August 2026. They also work with any MCP-compliant client (Continue, Warp, and others). The install commands vary slightly per client but the underlying server code is the same. See each server's dedicated page for client-specific setup instructions.
Yes with the right prompting. MongoDB and DynamoDB benefit hugely from having the MCP server connected because the AI can inspect indexes and previous document shapes before generating queries. Claude Opus 4.7 handles MongoDB aggregation pipelines well; GPT-5 is competitive. For Cassandra CQL, both providers can generate correct queries when they have access to the schema and can see the partition key structure — without that context, both struggle.

Hit an auth or connectivity error with a Database MCP?

Our sister sites cover errors, workarounds, and diagnostic recipes across the AI ecosystem. Bookmark them for the inevitable moment something breaks.

Visit AI Error Hub
SS
Curated by Skills and Servers Editorial
Editor-in-chief: Ahmed R. · Co-lead: Sana K.
Last updated August 17, 2026. Next full review of this category scheduled within 30 days. Every MCP server entry is installed, tested, and documented by a human before publication.
Suggest a server

Share with

Twitter/XPinterestRedditMastodonThreadsEmailPrintFlipboardFlipboardHacker NewsLineWeChatSnapchatMixMixPocketGoogle MessagesSMSTumblrTumblrXingXINGVKVKontakteBufferBufferYummlyNextdoorGrocyGabFacebookWhatsAppLinkedInTelegramMessengerBluesky