Skip to main content

Configure MCP

Amplifi allows you to integrate Model Context Protocol (MCP) tools to extend its core functionality.

🧠 MCPs let you run custom logic for parsing, enrichment, or external task execution inside your Amplifi workflows.

When you navigate to the MCP tab in the left navigation, you’ll see a built-in internal MCP tool called Tavily MCP Tool. This tool is already pre-configured by Amplifi and ready for immediate use.

In addition to this, you can add your own external MCPs for custom use cases.

Finding Existing External MCPs​

You can find existing external MCPs from several sources:

  • MCP Registry: Browse the official Model Context Protocol registry for community-contributed MCPs
  • GitHub: Search for MCP implementations on GitHub using tags like model-context-protocol or mcp-server
  • NPM: Look for MCP packages published on npm with the @modelcontextprotocol scope
  • Community Forums: Check MCP community discussions and documentation for recommended tools

Popular external MCPs include:

  • File System MCP: Browse workspace files, read document contents, and locate relevant assets for data analysis
  • Database MCP: Query structured datasets, execute SQL commands, and retrieve business intelligence from your connected databases

Adding an External MCP​

To add a new external MCP:

  1. Go to the MCP tab from the left navigation bar.

  2. Click Add External MCP on the top right.

  3. You’ll see a configuration form like this:

  4. Fill in the fields:

    • MCP Name – A short, unique identifier.
    • MCP Description – What this MCP does.
    • Paste Config Code – A JSON configuration block (see below).

Configuration Format​

To configure your external MCP, please use the following JSON format:

{
  "your_mcp_name": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/your-mcp-package", "arg1"],
    "env": {}
  }
}

### Notes

- Replace `"your_mcp_name"` with the identifier you wish to assign (e.g., `"custom_mcp"`).
- Replace `"@modelcontextprotocol/your-mcp-package"` with your actual MCP package.
- Modify `"arg1"` to the appropriate command required by your MCP.
- Use the `"env"` section to define any environment variables your MCP requires.

> ⚠️ Only MCPs that can be executed with the `npx` command are currently supported.