PhonePi MCP

Bridge desktop AI assistants to your phone over a local MCP server

Advanced API

What is PhonePi MCP?

PhonePi MCP is an AI mistress tool. An MIT-licensed MCP server plus Android app that lets Cursor, Claude Desktop and Cline send messages, place calls, manage contacts and push task notifications to your phone over a local network connection, with no third-party servers involved.

PhonePi MCP is a self-hosted bridge that lets desktop AI assistants act on an Android phone. It implements the Model Context Protocol, the same standard used by Cursor, Claude Desktop and Cline, and exposes a set of 23 phone actions that the model can call while it works. Rather than routing requests through a vendor cloud, the package runs an MCP server on the user's own machine and talks to a companion app on the phone across the local network, so contacts, snippets and notifications stay on the two devices. The practical appeal is task completion feedback. A developer who hands a long refactor or analysis job to Cursor no longer has to watch the screen; the assistant can fire a notification to the handset when the run finishes, with priority levels and optional response options for interactive confirmation. The same channel covers messaging, placing calls, adding and editing contact records, copying text to the phone clipboard, setting timers and beeping the handset to locate it. Snippets, the tool's note and bookmark store, can be created, searched, filtered by type, updated, deleted and shared back out through SMS or clipboard. Setup follows a short sequence: a JSON block added to the host app's MCP configuration launches the server with npx, the Android app is installed from Google Play and granted contacts, messages and notification permissions, and the phone is pointed at the computer's IP address on port 11041. Devices on different networks can be joined with Tailscale. A separate CLI wraps the same server for terminal use and accepts an Anthropic API key for natural language queries, and handles starting the server automatically. PhonePi MCP suits developers and power users who already work inside MCP-capable editors and want their handset to be a remote endpoint for those tools. It is not a general phone automation suite, and anyone wanting an iPhone or a hosted, zero-install experience should look elsewhere for now.

How do you use PhonePi MCP?

  1. 1Add the MCP server to your AI app config
    Create or edit the mcp.json style configuration for your client and add the phonepi-mcp entry that launches the server with npx and the start argument.
  2. 2Install the Android app and grant permissions
    Download PhonePi MCP from Google Play and allow the contacts, messages and notification permissions it requests so the AI tools can act on your behalf.
  3. 3Connect the phone to your computer
    Open the app and enter your computer's local IP address and port 11041. If the two devices sit on different networks, connect them first through Tailscale.
  4. 4Verify the connection status
    Confirm the green Connected indicator appears in both the mobile app and your IDE before sending any commands.
  5. 5Try a first query or the CLI
    Ask the assistant to beep your phone or check its battery, or run npx phonepi-cli query with an Anthropic API key for a one-off terminal request.
  6. 6Clear a stuck port when needed
    If the server will not start, run npx phonepi-mcp stop and free port 11041 with lsof or netstat before restarting from your AI app.

Pros and cons

Pros

  • Runs entirely on local hardware, with no third-party server handling phone dataAI
  • Works across any MCP-capable client, including Cursor, Claude Desktop and ClineAI
  • Broad tool set of 23 actions spanning messaging, calls, contacts, snippets and device checksAI
  • Notification and interactive message tools solve the real problem of unattended long AI runsAI
  • MIT-licensed implementation with an accompanying CLI for quick terminal queriesAI

Cons

  • Android only at launch; the iOS app is still listed as in developmentAI
  • Requires manual IP and port entry, so setup assumes comfort with local networkingAI
  • Only one AI app can bind port 11041 at a time, which blocks parallel assistantsAI
  • Messaging and call actions need broad phone permissions, a real trust decision for usersAI
  • Remote use outside the home network depends on a third-party mesh VPN such as TailscaleAI

How much does PhonePi MCP cost?

Support

Help runs through GitHub Discussions and an email contact listed on the homepage, with a troubleshooting section covering IP, port and network issues.

Integrations

Designed for MCP-capable desktop clients such as Cursor, Claude Desktop and Cline, with pairing through a JSON server configuration, an npx CLI for terminal use, and optional Tailscale for connecting devices across networks.

Features

Twenty-three callable tools cover SMS sending, phone calls, contact creation and editing, snippet storage with search and type filters, clipboard copy, push notifications with priority and response options, timers, battery checks and remote beeping to locate the handset. The server runs locally and the package is published under the MIT licence.

Frequently asked questions about PhonePi MCP

  • Does PhonePi MCP have an API?
    Yes, PhonePi MCP offers an API.
  • How do you use PhonePi MCP?
    The walkthrough on this page covers 6 steps: 1. Add the MCP server to your AI app config 2. Install the Android app and grant permissions 3. Connect the phone to your computer 4. Verify the connection status 5. Try a first query or the CLI 6. Clear a stuck port when needed.
  • What platforms does PhonePi MCP support?
    PhonePi MCP is available on Android, MacOS, Windows and Linux.
  • What does PhonePi MCP integrate with?
    Designed for MCP-capable desktop clients such as Cursor, Claude Desktop and Cline, with pairing through a JSON server configuration, an npx CLI for terminal use, and optional Tailscale for connecting devices across networks.
  • What are the limitations of PhonePi MCP?
    Android only at launch; the iOS app is still listed as in development. Requires manual IP and port entry, so setup assumes comfort with local networking. Only one AI app can bind port 11041 at a time, which blocks parallel assistants.