Bollard MCP

An AI-powered, safe database access layer built on the Model Context Protocol.

Bollard acts as an intelligent execution layer between AI code editors (Cursor, VS Code, Claude Desktop) and your databases — providing safe query execution, schema discovery, semantic intent validation, and persistent session memory.

Quick Start

pip install bollard-mcp

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "bollard": {
      "command": "bollard-mcp"
    }
  }
}

Then in Cursor chat:

Connect to my database: postgresql://user:pass@localhost/mydb
Show me all tables
Find users who signed up last week but haven't placed an order

Related MCP server: Polyglot DB MCP

Supported Databases

• PostgreSQL (primary)

• MySQL / MariaDB

• SQLite

• MS SQL Server

• Oracle

• MongoDB (Phase 2)

Key Features

• Safe by default — read-only mode, auto-LIMIT, WHERE enforcement, destructive operation blocking

• Intent validation — detects soft-delete blind spots, status ambiguity, missing tenant scope

• Session memory — schema cache, query history, user corrections persisted across restarts

• OS Keyring — credentials stored in system native vault (macOS Keychain, Windows Credential Manager, Linux Secret Service)

• No built-in LLM — editor AI handles reasoning; Bollard handles safe execution

Documentation

See docs/ for setup guides, permission profiles, and safety configuration.