PocketMCP: Turn Your Android Phone into an AI Agent Server

For all the incredible advancements in artificial intelligence over the past few years, there is still one glaring limitation: AI agents are usually trapped inside a browser tab or a desktop application.

They can write code, they can summarize PDFs, and they can search the web. But the moment you need your AI agent to interact with the real world—like texting a friend, checking your precise GPS location, or reading a 2FA code from an SMS—you hit a brick wall.

Your desktop doesn't have that data. Your phone does.

That’s why we built PocketMCP. It is a completely open-source Android application that securely turns your phone into an MCP (Model Context Protocol) server. By bridging your desktop AI environment with your mobile device over a local Wi-Fi connection, PocketMCP gives your AI agents "hands" and "eyes" in the physical world.

What is the Model Context Protocol (MCP)?

If you aren't familiar with it, the Model Context Protocol is an open standard introduced to standardize how AI models interact with external data sources.

Instead of writing a custom API integration for every single tool (Slack, Postgres, Jira), you simply run an MCP server. Your AI client (like Anthropic’s Claude Desktop or an AI-enabled IDE) connects to the server and instantly understands what tools and data are available.

Most MCP servers run on a PC. PocketMCP runs in your pocket.

The Architecture of a Mobile AI Server

PocketMCP acts as a WebSocket server directly on your Android device. It leverages Android’s native APIs to securely expose specific capabilities to the network.

Here is a look at what an AI agent can do once connected to PocketMCP:

1. Send and Read SMS: Ask your desktop AI to "Text Dave that I'm running 5 minutes late." The AI sends the payload to PocketMCP, which natively sends the SMS via your phone. No Twilio API keys required.
2. Access Telemetry: Retrieve live battery status, Wi-Fi signal strength, and environmental sensor data.
3. Fetch Location Data: Your desktop AI can ping your phone for its precise GPS coordinates to calculate a commute or provide hyper-local weather.
4. Notification Interaction: (Experimental) Allow scripts to read incoming Android notifications for 2FA codes or immediate alerts.

It changes the paradigm of AI development from "What web APIs can I call?" to "What hardware do I actually own?"

Security and Local Architecture

The absolute most important aspect of PocketMCP is security. Giving a desktop script access to your text messages is terrifying if done wrong.

That is why PocketMCP operates strictly over your local area network (LAN).

• It does not ping a centralized Axonix cloud server.
• It does not store your texts in an external database.
• It relies on mDNS (Multicast DNS) for local discovery.

When you launch the app, it broadcasts its presence on your home Wi-Fi. Your desktop script detects it locally and establishes a direct WebSocket connection. The traffic never leaves your house. Obviously, you must grant the Android app explicit permissions (like READ_SMS and ACCESS_FINE_LOCATION) before it can do anything, giving you granular control over what the MCP server exposes.

Why This Matters for the Future of AI

We are moving away from simple chatbots and toward autonomous AI agents. For an agent to be truly autonomous, it needs to interact with the devices you use every day.

By open-sourcing PocketMCP, we hope to accelerate the adoption of personal, localized AI infrastructure. You already own the hardware; your AI just needed the protocol to talk to it.

Get Involved

PocketMCP is completely free and completely open-source. Whether you are an Android developer who wants to contribute new MCP endpoints (like Camera access or Bluetooth scanning), or a desktop scripter who wants to build cool automations, the codebase is waiting for you.

Check out the source code, open an issue, or submit a pull request on our GitHub repository, and check out the Open Source page for our other projects.