Skip to Content
🚀 Orga AI is in open beta.
TutorialsMCP

MCP: Connect your first server

This tutorial walks you through connecting a local MCP server to Orga AI so your assistant can use external tools during a session.

What you will learn

  • What MCP is (at a practical level)
  • How to run a local MCP server (filesystem)
  • How to connect it to your Orga AI session with mcp_server

Prerequisites

  • Node.js 18+
  • An Orga AI integration that can create a WebRTC session
  • A local project directory you’re comfortable allowing the assistant to access

Only expose directories you are comfortable sharing with the AI assistant. MCP servers often provide powerful tools.

Run a filesystem MCP server

Start the MCP filesystem server and point it at a directory you want to allow access to.

npx @modelcontextprotocol/server-filesystem /path/to/allowed/directory

Leave this terminal running. Your Orga AI session connects to the server while it is running.

Confirm the SSE endpoint is reachable

By default, most MCP servers expose an SSE endpoint at /sse on port 3000.

curl http://localhost:3000/sse

This command should hang (it keeps the connection open waiting for server-sent events). If it exits immediately with an error, fix that before continuing.

Connect the MCP server in your session config

When you initialize or start a session, pass mcp_server pointing at the server’s SSE URL.

const sessionConfig = {
  // ...your existing session fields...
  mcp_server: {
    id: "filesystem",
    alias: "Local Files",
    url: "http://localhost:3000/sse",
  },
} as const;

Verify tools appear in the Orga UI

Start your session and look for the MCP server name (Local Files) in the UI. Once connected, the assistant should be able to use filesystem tools exposed by the server.

If the assistant says it cannot access files, jump to MCP troubleshooting.

Next steps

Last updated on
WidgetHow-to Guides