Postgres MCP
The gold standard for Postgres MCP servers. Explicit transactions, EXPLAIN output, and read-only mode make this the production-safe choice.
Database servers let models run queries, introspect schemas, and modify rows through the MCP protocol. The serious ones handle transaction boundaries, read-only modes, and schema introspection as first-class features. Use this page to separate them.
The gold standard for Postgres MCP servers. Explicit transactions, EXPLAIN output, and read-only mode make this the production-safe choice.
The production-grade database MCP. RLS-aware queries, schema management, and auth user handling make this the standard for agents that need structured persistence.
MindsDB is a different shape from every other server in this directory. Most MCPs are thin wrappers around one product's API; MindsDB is a query engine that federates 200+ data sources (Postgres, MongoDB, Slack, Salesforce, Shopify, files, and so on) and exposes them as a single SQL-compatible surface to the agent. The pitch is "give your agent access to all your live data through one connection," and the architecture delivers it: the agent issues SQL, MindsDB routes the query across the federation, results stream back. The cost is operational complexity. Running MindsDB means standing up a Docker container or PyPI install with a database backend, configuring connectors per data source, and managing a Knowledge Base for the unstructured side. Teams who already use MindsDB for analytics get the MCP integration nearly for free; teams considering MindsDB only for the agent surface should weigh the install footprint against using individual MCP servers per data source.
Official MCP server for PostgreSQL. Provides read and write access to PostgreSQL databases with schema introspection, query execution, and transaction support. The reference implementation for database MCP servers.
Persistence layer with full Postgres access and Row Level Security awareness. Query tables, manage schemas, handle auth users, and work with storage buckets.
A federated query engine that exposes 200+ data sources as MCP tools through a unified SQL-compatible interface. Built around a Connect → Unify → Respond workflow with structured tables fused with vectorized data inside Knowledge Bases.
MCP server for Neon serverless Postgres. Provides database branching, query execution, and project management. Branch-per-query makes every schema change reversible.
Vendor-built MongoDB MCP server covering both direct database operations (against any MongoDB connection string) and MongoDB Atlas API operations (via Service Accounts credentials). Ships with --readOnly enabled by default in every official install snippet.
Four Neo4j MCP servers maintained by the Neo4j Field GenAI team as part of Neo4j Labs: cypher (natural-language-to-Cypher with schema introspection), memory (knowledge-graph memory across agent sessions), cloud-aura-api (Aura cloud instance management), and data-modeling (graph data modelling and visualization). Each is a separate installable PyPI package.
MCP connects AI agents to external tools. The auth spec is solid. The attack surface around it is wide open. Here is what the protocol enforces, what it leaves to you, and how to close the gaps.
Most MCP tutorials are either too abstract or too long. Here is a working MCP server in Python in under 100 lines, exposing real tools, testable with Claude Code.
The navigation guide to MCP server discovery across 12,000+ servers. Compare PulseMCP, Smithery, Glama, MCP.so, the official registry, and every major index to find the right one for your workflow.
The 10 MCP servers that matter most for automation builders. Hands-on evaluation of each server's capabilities, setup friction, and production readiness.
We score a new database MCP the week it ships. Get the update, plus any re-scores when a server changes enough to move its rating. One email per week maximum.
// Tagged mcp-category-database · Unsubscribe any time