db-read

db-read is a read-only MCP server for coding agents that need safe access to PostgreSQL, MySQL, and MongoDB. It supports both hosted and local databases through a git-ignored .db-read.yml.

Features

• One MCP server for PostgreSQL, MySQL, and MongoDB
• Environment-based config: dev, staging, prod
• Local and hosted connection support via uri or explicit host settings
• Read-only guards for SQL and MongoDB aggregation
• MCP tools for listing connections, describing them, inspecting schema, and running read queries

Config

Copy .db-read.example.yml to .db-read.yml and fill in your credentials.

version: 1
defaultEnvironment: dev
defaults:
  timeoutMs: 10000
  maxRows: 1000
  maxDocuments: 1000
environments:
  dev:
    connections:
      app_pg:
        kind: postgres
        uri: postgresql://readonly:secret@localhost:5432/app
      billing_mysql:
        kind: mysql
        host: db.example.com
        port: 3306
        user: readonly
        password: secret
        database: billing
        ssl: true
      analytics_mongo:
        kind: mongodb
        uri: mongodb+srv://readonly:[email protected]
        database: analytics

CLI

npm run build
node dist/cli.js validate-config --config .db-read.yml
node dist/cli.js serve --env dev --config .db-read.yml

MCP client examples

Codex-style stdio config:

{
  "mcpServers": {
    "db-read": {
      "command": "node",
      "args": [
        "/absolute/path/to/db-read/dist/cli.js",
        "serve",
        "--env",
        "dev",
        "--config",
        "/absolute/path/to/project/.db-read.yml"
      ]
    }
  }
}

Cursor and Claude Code can use the same stdio command/args pattern.

Safety model

• Use least-privileged read-only DB credentials.
• SQL tools reject multi-statement and non-read queries.
• MongoDB tools expose only find and aggregation, and block $out and $merge.
• Timeouts and result limits are enforced per request.