---
title: "MCP for mobile developers, explained simply"
description: "MCP lets your AI agent see crash reports, Firebase data and Figma designs. Here's what it is, how Android Studio and Xcode use it, and how to stay safe."
author: "MH TOUFIK"
published: 2026-11-17T03:00:00.000Z
updated: 2026-11-13T16:52:36.000Z
url: https://toufik.dev/blog/mcp-for-mobile-developers
category: "Mobile AI"
tags: ["AI agents", "Android Studio", "Firebase", "MCP", "Xcode"]
---

# MCP for mobile developers, explained simply

If you've opened Android Studio's AI settings lately, or read Xcode 27's release notes, you've probably seen three letters: **MCP**. Most explanations of it are written for web or backend people, so mobile developers often skip it.

That's a mistake, because MCP is how your AI agent gets to see the things it can't see today: your crash reports, your Firebase data, your designs.

Here's what MCP is in plain words, why it matters for Android and iOS work, and how to start without opening a security hole.

## What MCP actually is

MCP stands for **Model Context Protocol**. Anthropic created it and released it as an open standard in **November 2024**. The official site describes it as a way of "connecting AI applications to external systems", and uses a nice comparison: **a USB-C port for AI apps**.

There are three parts:

- **The host** is the AI app you use, like Android Studio's agent, Xcode, or a CLI agent like Claude Code.
- **A server** is a small program that gives the AI access to one system, like Firebase or Figma.
- **A client** sits inside the host and keeps the connection to one server.

A server can offer three kinds of things. **Tools** are actions the AI can run, like "fetch the latest crashes". **Resources** are data it can read. **Prompts** are ready-made templates.

The big idea is simple. Before MCP, every AI tool needed its own plugin for every service. Now a service builds **one** MCP server, and any MCP-compatible host can use it.

![How MCP connects your AI to your tools: an AI app such as Android Studio, Xcode or Claude Code connects over MCP to servers like Firebase, Figma and your own tools, which offer tools, resources and prompts](https://toufik.dev/media/mcp-for-mobile-developers-122375a0 "One host, many servers. Each server gives your AI access to one system.")

## Why mobile developers should care

Both of our IDEs now speak MCP:

- **Android Studio:** Google's docs say the Gemini agent "can interact with external tools using the Model Context Protocol". I covered how Android Studio is opening up to other agents in my [Android Dev Summit 2026 post](https://toufik.dev/blog/android-dev-summit-2026-ai-agents-android-studio).
- **Xcode 27:** Apple says you can "extend Xcode with your own skills and tools using the Model Context Protocol."

So the same server works in both. Set up a Figma server once, and your agent can read the same design whether you're writing Compose or SwiftUI.

## Two servers worth trying first

**1. Firebase.** Google's official Firebase MCP server runs on your machine and gives the agent tools for **Crashlytics, Authentication, Firestore and Cloud Messaging**, among others. That means you can ask "what are the top crashes in the last release, and where in the code do they start?" and the agent can actually look.

With Claude Code, adding it is one line:

```bash
claude mcp add firebase npx -- -y firebase-tools@latest mcp
```

**2. Figma.** Figma's remote MCP server lives at `https://mcp.figma.com/mcp`, so you don't even need the Figma desktop app. The agent can read real spacing, colours and components from a frame, instead of guessing from a screenshot. Figma's docs say Xcode supports this remote server from the **Xcode 27 beta**.

My advice: start with **one** server that fixes a real daily pain. For most mobile teams, that's crash reports.

## Adding a server to Android Studio

In Android Studio, open **Settings → Tools → AI → MCP Servers** and turn on **Enable MCP Servers**. The setup is saved in an `mcp.json` file, like this:

```json
{
  "mcpServers": {
    "figma": {
      "httpUrl": "https://mcp.figma.com/mcp"
    }
  }
}
```

One thing to know: Google's page lists a few limits. Android Studio **doesn't support the `stdio` transport** yet, so local servers that run as a command, like the Firebase one, won't work there. It also doesn't support MCP resources or prompts yet. For local servers, a CLI agent next to your IDE works fine.

## Be careful what you plug in

This is the part people skip, and it's the most important one.

The MCP spec says it plainly: **"Tools represent arbitrary code execution and must be treated with appropriate caution."** An MCP server can read your data and act for you. A bad one can do real damage.

OWASP now has an **MCP Top 10** list of risks. In simple words, the big ones are:

- **Leaked tokens.** API keys sitting in config files or logs.
- **Too many permissions.** A server that can delete when it only needs to read.
- **Tool poisoning.** A server whose tool descriptions secretly tell the AI to do something else.
- **Prompt injection.** Text in a crash report, ticket or web page that tries to give your agent orders.
- **Shadow servers.** Servers someone on the team added that nobody reviewed.

My rules:

1. **Only install official servers**, or ones whose code you've read.
2. **Give read-only access** wherever you can.
3. **Keep tokens out of your repo.** Use environment variables.
4. **Review every action** before the agent runs it, especially anything that writes or sends.
5. **Remove servers you don't use.** Fewer servers, fewer surprises.

## My take

MCP isn't magic, and it won't write your app for you. What it does is give your agent the same context you already have: the crash, the design, the data. That's where most AI mistakes come from today, guessing without context.

Start with one trusted server, keep it read-only, and see how much less you copy and paste in a week.

## Sources

- [What is the Model Context Protocol?](https://modelcontextprotocol.io/docs/getting-started/intro) — Model Context Protocol
- [Architecture overview](https://modelcontextprotocol.io/docs/learn/architecture) — Model Context Protocol
- [Introducing the Model Context Protocol](https://www.anthropic.com/news/model-context-protocol) — Anthropic
- [Add an MCP server](https://developer.android.com/studio/gemini/add-mcp-server) — Android Developers
- [Xcode](https://developer.apple.com/xcode/) — Apple Developer
- [Firebase MCP server](https://firebase.google.com/docs/ai-assistance/mcp-server) — Firebase
- [Figma remote MCP server](https://developers.figma.com/docs/figma-mcp-server/remote-server-installation/) — Figma
- [OWASP MCP Top 10](https://owasp.org/www-project-mcp-top-10/) — OWASP
