
CentralMind Gateway
OfficialAuto-generate a secure, LLM-optimized MCP server or REST API for your database in minutes.
Add to your client
Copy the config for your MCP client and paste it into its config file.
docker run --platform linux/amd64 ghcr.io/centralmind/gateway:v0.2.18 startPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"centralmind-gateway": {
"command": "PATH_TO_GATEWAY_BINARY",
"args": [
"start",
"--config",
"PATH_TO_GATEWAY_YAML_CONFIG",
"mcp-stdio"
]
}
}
}Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- A supported database (PostgreSQL, MySQL, ClickHouse, Snowflake, MSSQL, BigQuery, Oracle, SQLite, or ElasticSearch) with a connection string
- The gateway binary (built from source with Go, or run via the ghcr.io/centralmind/gateway Docker image)
- An AI provider API key for the discovery stage (OpenAI, Anthropic, Amazon Bedrock, Google Gemini, or Google Vertex AI) — e.g. GEMINI_API_KEY
- A generated gateway.yaml configuration file (produced by the `gateway discover` command)
About CentralMind Gateway
CentralMind Gateway connects to a structured database and auto-generates a secure, LLM-optimized API exposed over MCP (stdio or SSE) and/or REST/OpenAPI 3.1. An LLM is used only during a one-time discovery stage to analyze the schema and produce a gateway.yaml config in which each endpoint becomes an MCP method. It adds PII redaction, authentication, row-level security, caching, and OpenTelemetry auditing on top.
Tools & capabilities (1)
<dynamically generated per endpoint>Gateway does not expose a fixed set of MCP tools. During the `discover` stage, an LLM generates a gateway.yaml in which each database endpoint is given an `mcp_method` (with summary, description, SQL query, and params). Those generated methods become the MCP tools the server exposes — for example read-only query methods over your tables. The exact tool names and descriptions depend on your database schema and the discovery prompt.
When to use it
- Chatting with your database or data warehouse from Claude Desktop, Cursor, or other AI agents
- Letting an LLM create, adjust, or query data during development via remote function/tool calling
- Exposing a structured database as a secure, AI-optimized REST or MCP API with PII protection and auditing
- Enriching AI agents with live data from PostgreSQL, Snowflake, BigQuery, and other supported databases
Security notes
The MCP server itself does not require client-side authentication in the documented stdio config; the gateway binary connects to your database using a connection string in gateway.yaml. Gateway provides PII filtering/redaction (regex or Microsoft Presidio), API key and OAuth auth for its REST/SSE endpoints, row-level security via Lua scripts, and OpenTelemetry auditing. The README notes enhanced MCP-level API key/OAuth authentication is still on the roadmap.
CentralMind Gateway FAQ
How do I add Gateway to Claude Desktop?
Add an entry to Claude Desktop's config under mcpServers with command set to the path of the gateway binary and args ["start", "--config", "PATH_TO_GATEWAY_YAML_CONFIG", "mcp-stdio"]. You must first generate the gateway.yaml config using the `gateway discover` command.
What MCP tools does the server expose?
There is no fixed tool list. Gateway uses an LLM during the discovery stage to generate an API configuration where each endpoint maps to an mcp_method. Those generated methods are the tools exposed at runtime, so they depend entirely on your database schema and discovery prompt.
Does it support transports other than stdio?
Yes. Besides the stdio MCP server used by Claude Desktop, Gateway can run an MCP SSE server for AI agents (e.g. http://localhost:9090/sse) and a REST API with Swagger UI / OpenAPI 3.1.
Is my data protected?
Gateway includes PII detection and redaction (via a regex plugin or Microsoft Presidio), API key and OAuth authentication for its endpoints, row-level security through Lua scripts, and OpenTelemetry-based request tracking and audit trails.
Alternatives to CentralMind Gateway
Compare all alternatives →Google's official MCP server with prebuilt BigQuery tools, querying datasets via Application Default Credentials.
Read/write Postgres access plus index tuning, EXPLAIN plans, and database health analysis for AI agents.
Official Supabase server: manage tables, run SQL, branches, configs and edge functions from your AI client.
Compare CentralMind Gateway with: