MCP SQL Bridge

MCP SQL Bridge

Local-SQL-Insight — Give your AI assistant direct, read-only access to your SQL databases over the Model Context Protocol. No APIs, no cloud. Your data stays on your machine.

Why use this?

• Schema-aware AI — The agent sees table names, columns, and full CREATE TABLE DDL so it can write correct SQL.
• Business context — Expose a project README.md as a resource so the AI understands your domain and conventions.
• Read-only by design — Only SELECT is allowed. No accidental writes or schema changes.
• Stdio transport — Works with Cursor, Claude Desktop, and any MCP host. No ports or network setup.
• Multi-database (Pro) — Use the same tools against SQLite, PostgreSQL, and MySQL by switching a backend and connection string.

Quick start

git clone <your-repo-url>
cd mcp-sql-bridge
poetry install

Then plug the server into your editor or desktop app (see below).

Run & test: See RUN_AND_TEST.md for how to run the MCP server, the web server, unit tests, and Docker-based integration tests (e.g. poetry run python scripts/run_docker_integration.py).

Web server (optional)

A small FastAPI app provides a landing page, Swagger UI, and documentation (MkDocs):

poetry install --with web
mkdocs build          # build docs (optional; without this, /documentation shows build instructions)
poetry run mcp-sql-bridge-web

Open http://localhost:8000 for the home page, http://localhost:8000/docs for Swagger, http://localhost:8000/redoc for ReDoc, and http://localhost:8000/documentation for the full docs site (after mkdocs build).

How to use with Cursor

1. Install the server (if not already):

   cd /path/to/mcp-sql-bridge
   poetry install
   

2. Configure MCP Open Cursor Settings → Tools & MCP (or edit the config file directly). Add the server using the template in mcp-config.json in this repo:

   • Option A — Global config Put the contents of mcp-config.json into:

     • macOS / Linux: ~/.cursor/mcp.json
     • Windows: %APPDATA%\Cursor\mcp.json

   • Option B — Project config Copy mcp-config.json into your project as .cursor/mcp.json (and adjust the cwd path to this repo if needed).

3. Set the path In the config, set cwd (or the path in args) to the absolute path of your mcp-sql-bridge project directory so Cursor can run poetry run mcp-sql-bridge from there.

4. Restart Cursor MCP servers load at startup. Restart Cursor after changing the config.

5. Use it In any chat, the AI can call Local-SQL-Insight tools: list_tables, execute_readonly_query, and read the Database README resource.

   • For SQLite, point it at a file path (e.g. ./data/app.db) with backend: "sqlite" or by omitting backend.
   • For PostgreSQL/MySQL (Pro), provide a backend and connection_string URL.

How to use with Claude Desktop

1. Install the server (if not already):

   cd /path/to/mcp-sql-bridge
   poetry install
   

2. Locate Claude’s config file

   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json

3. Add the MCP server Add a mcpServers entry (or merge into existing) like this, using the absolute path to your mcp-sql-bridge directory:

   {
     "mcpServers": {
       "local-sql-insight": {
         "command": "poetry",
         "args": ["run", "mcp-sql-bridge"],
         "cwd": "/absolute/path/to/mcp-sql-bridge"
       }
     }
   }
   

   If Claude Desktop uses a different Python environment, you can instead use the full path to the mcp-sql-bridge executable in the project’s virtualenv, for example:

   {
     "mcpServers": {
       "local-sql-insight": {
         "command": "/absolute/path/to/mcp-sql-bridge/.venv/bin/mcp-sql-bridge",
         "args": []
       }
     }
   }
   

4. Restart Claude Desktop Restart the app so it picks up the new MCP config.

5. Use it Claude can use the list_tables and execute_readonly_query tools and read the Database README resource.

   • For SQLite, use your SQLite file path (e.g. ~/project/data/app.db).
   • For PostgreSQL/MySQL, supply connection URLs when the model asks for database details.

What the server provides

For SQLite, the server validates that paths point to real SQLite files (magic header check) and returns clear errors if a path is wrong or not a database. For PostgreSQL/MySQL, it validates connections and reports connection/auth errors clearly; use read-only roles in production.

Connection strings (PostgreSQL & MySQL)

Use the connection_string argument with backend: "postgres" or backend: "mysql" when calling list_tables or execute_readonly_query.

PostgreSQL — URL format (password in URL is optional; use env vars in production):

• postgresql://user:password@localhost:5432/dbname
• postgresql://user@localhost:5432/dbname (no password)
• postgresql://localhost:5432/dbname (peer/auth trust)

MySQL — URL format:

• mysql://user:password@localhost:3306/dbname
• mariadb://user:password@localhost:3306/dbname

Install optional drivers: poetry install --extras postgres and/or poetry install --extras mysql (or --extras pro for both). Example MCP configs for multi-database setups are in examples/mcp-config-pro.json and examples/README.md.

Documentation

• Main README (this file) — quick start and usage.
• docs/ — security and hardening, error taxonomy, pricing and licensing, launch and marketing, future ideas.
• CONTRIBUTING.md — development setup, tests, release, PyPI.

Requirements

• Python 3.11+
• Poetry (recommended) or pip for dependencies
• Optional extras:
  • postgres extra for PostgreSQL (psycopg)
  • mysql extra for MySQL (pymysql)

Feedback and support

• Bug reports and feature ideas: Open an issue.
• Questions and discussions: GitHub Discussions.
• Feedback from users (issues, discussions, or contact) is used to prioritize the roadmap and any future paid offerings.

Support & maintenance

• Targeted Python: 3.11+
• MCP: current stable versions at time of release
• Backwards-compatible changes will be released as minor versions; breaking changes will bump the major or minor version and be noted in CHANGELOG.md.
• The project is intended to be feature-complete after the 0.2.x line; future work will focus on fixes and compatibility updates only.

License

MIT. See LICENSE for details.