MCP Server

SoftDB includes a built-in MCP (Model Context Protocol) server that lets AI coding assistants — Claude Desktop, Cursor, Windsurf, and others — directly browse your database schemas, execute queries, and explore relationships. No copy-pasting schemas into chat. The AI just knows your database.

Quick Setup

Pick your AI tool and paste the config:

Claude Desktop

Add to ~/.config/claude/claude_desktop_config.json (macOS/Linux) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "softdb": {
      "url": "http://localhost:9090/mcp"
    }
  }
}

Restart Claude Desktop after saving.

Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "softdb": {
      "url": "http://localhost:9090/mcp"
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "softdb": {
      "url": "http://localhost:9090/mcp"
    }
  }
}

Stdio (any MCP client)

For MCP clients that support stdio transport:

{
  "mcpServers": {
    "softdb": {
      "command": "/path/to/softdb-mcp"
    }
  }
}

Build the binary: go build -o softdb-mcp ./cmd/mcp/

Tip

The HTTP server runs on port 9090 by default. You can change it in Settings → MCP Server → Server Port.

Enabling the MCP Server

1. Open SoftDB and go to Settings (gear icon in the toolbar).

2. Click MCP Server in the sidebar.

3. Toggle Enable MCP Server on. The status indicator turns green and shows the active port.

4. Copy the config snippet shown in the settings panel and paste it into your AI tool’s MCP config file.

5. Restart your AI tool. It will connect to SoftDB automatically.

Available Tools

Once connected, the AI agent has access to 8 database tools:

Tool	Description
list_connections	List all MCP-enabled database connections
use_connection	Set the active connection (auto-connects if needed)
list_databases	List databases on the active connection
list_tables	List tables, optionally in a specific database
describe_table	Get column definitions — name, type, nullable, PK, default
execute_query	Run SQL, MongoDB JSON queries, or Redis commands
read_table	Paginated table data browsing
get_relationships	Foreign key relationships between tables

Per-Connection Access Control

Not every database should be visible to AI. SoftDB lets you control which connections the MCP server can access:

• In the Connection Hub, hover over a connection card to reveal the MCP icon (hub icon).
• Click it to toggle MCP access for that connection.
• Disabled connections are hidden from list_connections and cannot be used by AI agents.

All connections are MCP-enabled by default.

Safety

MCP inherits SoftDB’s per-connection Safe Mode:

• When Safe Mode is on, execute_query blocks destructive operations — DROP, TRUNCATE, DELETE without WHERE, UPDATE without WHERE.
• The AI agent receives a clear error explaining why the query was blocked.
• Query results are capped at 1000 rows to prevent oversized responses.
• Credentials are never exposed — the AI only sees connection names, types, and hosts. Passwords, SSH keys, and URIs are stripped from all responses.

Caution

The MCP server listens on localhost only and has no authentication. Do not expose the port to the network. This is designed for local AI tools running on the same machine.

Two Transport Modes

HTTP (in-app)

When the SoftDB app is running with MCP enabled, it serves an HTTP endpoint on http://localhost:{port}/mcp. AI tools connect via HTTP. You control it from the Settings UI — toggle on/off, change port, see status.

Stdio (headless)

For CI/automation or when you don’t want the GUI, build and run the standalone binary:

go build -o softdb-mcp ./cmd/mcp/

Configure your AI tool to spawn it as a subprocess:

{
  "mcpServers": {
    "softdb": {
      "command": "./softdb-mcp"
    }
  }
}

The stdio binary reads your saved connections from the same config directory as the app (~/.config/SoftDB/).

Example Conversation

Once MCP is connected, you can ask your AI assistant things like:

“Show me all tables in the production database”

The AI calls use_connection → list_tables and returns the result.

“What columns does the users table have?”

The AI calls describe_table with table: "users" and formats the schema.

“Find all orders placed in the last 7 days with total > $100”

The AI calls execute_query with the appropriate SQL, gets the results, and presents them.

“How are the orders and products tables related?”

The AI calls get_relationships and explains the foreign key structure.