MCP Extensions

MCP (Model Context Protocol) is an open protocol that lets AI connect to external tools. Through MCP, you can give SailFish’s AI Agent direct access to query databases, call APIs, and operate various external services—like installing dedicated plugins for the AI.

What is MCP

Without MCP

When the AI needs to query a database, it must do so indirectly through the terminal:

You: How many users are in the users table?
AI: [Runs mysql -u root -p -e "SELECT COUNT(*) FROM users"]
    → Needs password, handles command-line output parsing...

With MCP

The AI operates directly through structured interfaces:

You: How many users are in the users table?
AI: [Calls the MySQL MCP tool's query method]
    → Gets structured results directly, faster and more accurate

The MCP ecosystem includes many open-source tools covering databases, file systems, version control, cloud services, and more.

Adding MCP Tools

SailFish includes preset templates for popular MCP tools:

  1. Open Settings → MCP Configuration
  2. Click Add
  3. Select the tool you need from the template list (e.g., MySQL, PostgreSQL)
  4. Fill in the required configuration (e.g., database address, username, password)
  5. Save and enable

Manual Configuration

If no preset fits your needs, you can manually configure any MCP-compatible tool:

  1. Click Add → Manual Configuration
  2. Fill in the configuration:
Field Description Example
Name Custom name for identification “Production Database”
Transport stdio (local process) or SSE (remote service) stdio
Command Command to start the MCP connector (stdio mode) npx @modelcontextprotocol/server-mysql
Arguments Command-line arguments --host 10.0.1.50 --port 3306
Environment variables Variables passed to the MCP process MYSQL_PASSWORD=xxx
URL MCP connector address (SSE mode) http://localhost:8080/sse

Two Transport Modes

Transport How It Works Use Case
stdio SailFish starts a local process and communicates via stdin/stdout Most scenarios (recommended)
SSE Connects to a remote HTTP server via Server-Sent Events MCP connector hosted on a remote server
MCP Tool Function Use Case
mysql Query and operate MySQL databases Data analysis, DB operations
postgres Query and operate PostgreSQL Data analysis, DB operations
sqlite Operate SQLite databases Local data file analysis
filesystem Enhanced file system operations Complex file management
git Git repository operations (commit, branch, log) Code management and review
fetch Send HTTP requests REST API calls
puppeteer Browser automation Screenshots, web scraping
memory Persistent knowledge graph Storing structured information
everything Windows file search Quick file lookups

Browse the full MCP tool catalog at MCP official directory.

Using MCP Tools

Once MCP is configured, the AI automatically detects available tools and invokes them when needed. Just describe what you want in natural language:

Database Queries

How many users registered in the last 7 days?
What's the total order amount for today in the orders table?
List products with stock below 10 in the products table

API Calls

Call our internal API and get the current number of online users
Search my entire computer for Word files with "report" in the filename

Managing MCP Tools

From the MCP configuration section in Settings you can:

Action Description
Enable/Disable Temporarily turn off an MCP tool (without removing config)
Reconnect Reconnect when the MCP process unexpectedly disconnects
View tool list See which tools and parameters the MCP provides
Edit Change configuration (e.g., database address)
Delete Remove MCP configurations you no longer need

Troubleshooting

MCP Tool Fails to Start

  • Ensure the required npm package is installed (most MCP tools run via npx)
  • Check that your Node.js version meets requirements (18+ recommended)
  • Check SailFish logs for error details

Database MCP Cannot Connect

  • Verify database address, port, username, and password
  • Ensure the database allows connections from your machine
  • Check firewall rules for the database port

MCP Tool Not Used by AI

  • Confirm MCP status is “Connected”
  • Try explicitly mentioning the feature in your prompt (e.g., “query the database”)
  • Check the MCP tool list to confirm it provides the capability you need