AI assistant and MCP
Ask about your database with the AI provider you choose, and let AI clients work with saved connections through MCP (Model Context Protocol).


Choose your AI provider
Use GitHub Copilot, ChatGPT, Cursor, Claude, Claude Agent, OpenAI, OpenRouter, Gemini, xAI, Ollama, llama.cpp, MLX, OpenCode Zen and OpenAI-compatible with an API key, an account or a local model. Requests go straight to the provider; TablePro runs no AI service. The assistant is free, and your provider may charge.
From the editor, review, explain or optimize a statement, or fix one that failed. Nothing runs, and a suggestion reaches the editor only when you apply it.
Result rows stay out of messages unless you allow them in Settings > AI. Linux supports Anthropic, OpenAI, Ollama and OpenAI-compatible.
Chat modes and Agent mode
Ask, the default, reads the schema and runs nothing. Edit runs queries, and Agent can also propose destructive statements, under the connection’s Safe Mode: Alert and stricter levels confirm writes, Read-Only refuses them. Full-window Agent mode keeps the connection at Alert or stricter while it works.
Connect AI clients through MCP
The built-in MCP server lets clients such as Claude Code, Claude Desktop, Cursor and Zed use your saved connections without seeing their passwords. It is off until you turn it on, listens on 127.0.0.1 only, and logs every call.
Each call passes the token’s scope, the connection’s External Clients level, its approval setting and Safe Mode, and the strictest one wins.
On Linux the server offers 9 tools, over HTTP.


Also
External MCP tools
The assistant can call an outside MCP server over Streamable HTTP. Every tool call waits for your click.
Automation
AppleScript,
tablepro://links and the Raycast extension open connections and run queries, within each connection’s limits.Where your data goes
Prompts and schema go to your provider under its terms, or stay on your machine with a local model. TablePro receives none of it.
Get TablePro
Everything on this page is free.