Gateway

What centralmind/gateway does

The centralmind/gateway MCP server exposes structured database data to AI agents through Model Context Protocol or generated REST APIs. It can also publish an OpenAPI 3.1 specification and Swagger documentation for the generated interface. The project is intended for development workflows where an LLM needs to query or modify database data, analytical use cases such as chatting with a warehouse, and agent applications that call database-backed functions.

Supported database connectors listed by the project include PostgreSQL, MySQL, ClickHouse, Snowflake, BigQuery, MSSQL, Oracle Database, SQLite, and Elasticsearch. The generated interface can be served through MCP, MCP over SSE, REST, or direct and raw SQL API modes.

How it works

At startup, the gateway connects to a database, reads its schema, and examines sampled data. A configured LLM provider uses that information and a supplied prompt to create an API configuration designed for the available data. According to the project documentation, the LLM is used during discovery rather than for every database request.

The resulting service can run locally, in Docker, or in Kubernetes. A typical Docker deployment listens on port 9090, exposes an MCP SSE endpoint at /sse, and provides a REST interface with Swagger UI at the root URL. Clients can then connect through MCP or call the generated HTTP API.

Setup and configuration

The repository can be built from source with Go by cloning the project, downloading its modules, and running go build .. The README also demonstrates a Docker image hosted at ghcr.io/centralmind/gateway, started with the start subcommand and a database connection string.

API generation requires configuration for one of the documented AI providers: OpenAI-compatible services, Anthropic, Amazon Bedrock, Google Vertex AI through Anthropic, or Google Gemini. Database connection details are supplied through the gateway configuration or startup arguments.

YAML configuration and plugins control customization. Deployment documentation covers standalone binaries, Docker, and Kubernetes. The project also documents integrations with Claude Desktop, Cursor, LangChain, and OpenAI-based applications.

Tools and capabilities

• Generates an API from database schema information and sampled records.
• Serves MCP, including an SSE transport, alongside REST endpoints.
• Produces OpenAPI 3.1 output and Swagger documentation.
• Supports API-key and OAuth authentication options.
• Provides PII redaction through a regular-expression plugin or Microsoft Presidio.
• Supports Lua-based row-level security rules.
• Offers time-based and LRU caching options.
• Emits request telemetry and audit information through OpenTelemetry.
• Can use hosted AI providers or configurable self-hosted model endpoints.

Limitations and notes

The centralmind/gateway MCP server depends on access to a supported database and an AI provider for API discovery. Security controls such as PII filtering, authentication, row-level security, and telemetry are available as configuration features; they still need to be configured for the deployment.

The generated API reflects the schema and sampled data available during discovery. Changes to database structure or desired access rules may require regenerating or updating the configuration. The project documentation should be consulted for connector-specific options and deployment details.