Let AI Agents Query Your Database Without Writing Endpoints

What it does

CentralMind Gateway connects to structured databases—PostgreSQL, MySQL, Snowflake, and others—and automatically produces REST and MCP endpoint configurations by analyzing schema and sampled data. An LLM acts as a one-time API designer during discovery, but actual queries run directly against your database with no further AI involvement. The generated gateway exposes your data to tools like Cursor, Claude Desktop, and LangChain through standard protocols.

The interesting bit

Instead of using an LLM to translate natural language into SQL on every request, the project generates a static YAML configuration up front. That keeps latency predictable and prevents live database traffic from reaching model providers. It also layers on PII redaction, row-level security via Lua scripts, and OpenTelemetry tracing, treating “AI access to production data” as a compliance problem, not just an integration one.

Key highlights

• Broad database support: PostgreSQL, MySQL, ClickHouse, Snowflake, BigQuery, Oracle, SQLite, and Elasticsearch.
• Protocol flexibility: serves both OpenAPI 3.1 REST APIs and MCP servers (including SSE mode).
• Security and compliance plugins: PII removal via regex or Microsoft Presidio, API keys, OAuth, and Lua-based row-level security.
• Observability and performance: OpenTelemetry integration plus LRU and time-based caching.
• Local-friendly: discovery can run against self-hosted LLMs so schema and sample data never leave your network.

Caveats

• Write support is still on the roadmap; current examples and tooling focus heavily on read-only access.
• Discovery inherently shares schema structure and data samples with the configured LLM provider unless you switch to a local model.
• Several connectors and enterprise features—such as Databricks, Redshift, and SSH tunneling—remain planned but not yet shipped.

Verdict

Data teams looking to expose existing SQL stores to AI agents through MCP or REST will find a useful head start here. If you need mature write APIs, complex transactions, or deep connector coverage beyond the current set, it is worth monitoring until the roadmap catches up.