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.

Real-world use cases we've seen

Since releasing PocketMCP, the community has come up with some creative applications that we didn't anticipate:

Morning briefing automation. One developer built a script that runs every morning. It connects to PocketMCP, checks the phone's battery level, reads the weather from a notification, grabs the calendar events from the next notification, and generates a morning briefing in the terminal. The AI agent formats it into a summary with suggestions for the day.

SMS-based code review. A team lead set up a workflow where a Claude Desktop agent monitors a GitHub repository. When a PR is approved, it sends an SMS through PocketMCP to the developer's phone with a summary of what was approved and which branch to merge. No Slack dependency, no email required.

Location-aware reminders. A freelancer connected their GPS data to a desktop automation script. When the phone reports that the user is near a specific location (detected via coordinates), the script triggers a reminder. The phone provides the location, the desktop provides the intelligence.

Remote device monitoring. Someone built a dashboard that periodically pings PocketMCP on three phones in their household. It displays battery levels, signal strength, and the last seen timestamp for each device. Useful for keeping track of whether the kids' phones need charging before they leave the house.

What PocketMCP can't do (yet)

We want to be honest about the current limitations:

No camera access. Android's camera API requires a foreground activity, which makes it impractical for a background MCP server. We're exploring workarounds, but it's not ready yet.

No file system access. For security reasons, PocketMCP doesn't expose the phone's file system. The risk of a malicious script reading personal files is too high. If there's demand, we might add a sandboxed photo gallery endpoint.

No persistent background service. Android kills background services aggressively to save battery. PocketMCP runs as a foreground service with a persistent notification, which keeps it alive but drains battery faster than a normal app. For always-on scenarios, you'll want to keep the phone plugged in.

Wi-Fi only. PocketMCP requires both devices to be on the same local network. It doesn't work over cellular or the internet. This is intentional for privacy, but it means you can't use it when your desktop and phone are on different networks.

How to get started

1. Install PocketMCP from the GitHub releases.
2. Make sure your phone and computer are on the same Wi-Fi network.
3. Open the app and grant the permissions you want to expose (SMS, location, notifications).
4. In your MCP client (Claude Desktop, Cursor, or any MCP-compatible tool), add PocketMCP as a server using the WebSocket URL shown in the app.
5. Start asking your AI agent to interact with your phone.

The whole setup takes about five minutes if you already have an MCP client configured. The app is free, open-source, and has open-source and free to use.

Security considerations for real-world use

Giving an AI agent access to your SMS and location sounds terrifying. Here is how to think about it safely.

Start with read-only permissions. Battery level and signal strength are harmless to expose. Try those first. Get comfortable with how the WebSocket connection works before enabling SMS or location access.

Use a dedicated Wi-Fi network. PocketMCP runs on your local network, so anyone on the same Wi-Fi could theoretically discover the server. Use your home network, not a coffee shop or hotel network.

Revoke permissions you do not use. If you only need SMS access for a specific automation, grant it, run the automation, and revoke it afterward. The Android permission system makes this straightforward.

Monitor the WebSocket traffic. Open your browser's DevTools network tab while PocketMCP is running. You will see the exact data being exchanged between your desktop and your phone. Nothing is hidden. If you see data you did not expect, investigate before continuing.

The MCP protocol itself does not add encryption beyond what your Wi-Fi network provides. For most home use, this is fine. If you need encryption, run a VPN or SSH tunnel between your desktop and phone.