getsentry/MobileBuildMCP

What it is

MobileBuildMCP is a small open‑source package that runs a Model Context Protocol (MCP) server and a matching command‑line interface (CLI). Its purpose is to give AI coding agents (e.g., Claude, Cursor, Codex) a way to invoke Xcode‑related build and test commands for iOS/macOS projects without the agent having to know the exact xcodebuild syntax.

Who might use it

• AI‑powered development assistants that need to compile, test, or package an iOS/macOS app as part of an automated workflow.
• Human developers who want a convenient CLI to run common Xcode build tasks (simulator builds, test‑product preparation, device signing) from a single, cross‑platform entry point.
• Tooling teams building higher‑level automation on top of Xcode and wanting a stable, language‑agnostic RPC surface (the MCP server).

Core features (as described in the README)

Feature	What it does
MCP server	Listens for requests from MCP‑compatible AI agents and exposes a set of Xcode‑related tools (build, test, device signing, etc.) over the Model Context Protocol.
CLI	Provides the same tools directly from the terminal (mobilebuildmcp <subcommand>).
Two‑mode install	Available via Homebrew (brew install mobilebuildmcp) or as a global npm package (npm install -g mobilebuildmcp).
Agent skills	Optional “MCP Skill” and “CLI Skill” JSON blobs that can be fed to an AI agent to teach it how to talk to the server or use the CLI.
Simulator workflow	Commands to build for a specific simulator, optionally produce test bundles without running them, and later execute those bundles on a chosen simulator.
Device tooling	Supports building for real devices, assuming code‑signing is set up in Xcode.
Upgrade helper	mobilebuildmcp upgrade --check / --yes lets users update the installed binary in place.
Telemetry	Sends runtime error data to Sentry (opt‑out possible).
Documentation	Extensive markdown docs hosted in a companion repo (getsentry/xcodebuildmcp.com) covering installation, configuration, tool reference, and troubleshooting.

How it works (high‑level)

1. Installation – The user installs the package (Homebrew or npm). The install provides both the mobilebuildmcp binary and the MCP server code.
2. Starting the server – Running mobilebuildmcp mcp launches an HTTP‑like MCP endpoint that listens for JSON‑RPC style requests from AI agents.
3. Agent interaction – An AI agent loads the provided MCP Skill (or CLI Skill) which tells it which JSON methods are available (e.g., simulator.build, simulator.test). The agent then sends a request, the server translates it into the appropriate xcodebuild command, runs it, captures logs, and returns the result.
4. CLI fallback – If the user prefers manual control, they can invoke the same operations directly via subcommands such as mobilebuildmcp simulator build ….
5. Stateful daemon – The CLI spins up a per‑workspace background daemon to keep logs, cache state, and speed up repeated operations.

Why it matters for AI‑assisted development

• Standardised interface: By exposing Xcode functionality through MCP, any LLM‑based coding assistant can interact with iOS/macOS projects without custom scripting for each tool.
• Reduced prompt complexity: The skill files give the agent a concise “API spec”, letting prompts stay focused on business logic rather than low‑level build flags.
• Cross‑platform entry point: Developers can run the same commands locally or let an AI agent run them remotely, making it easier to integrate into CI/CD pipelines that already use AI‑driven code generation.

Getting started (quick steps)

# Homebrew (recommended on macOS)
brew tap getsentry/xcodebuildmcp
brew install mobilebuildmcp

# Or npm
npm install -g mobilebuildmcp@latest

# Verify
mobilebuildmcp --help

# Start the MCP server for an AI client
mobilebuildmcp mcp

# Example CLI build for a simulator
mobilebuildmcp simulator build \
  --scheme MyApp \
  --project-path ./MyApp.xcodeproj \
  --simulator-name "iPhone 17"

License & community

• Licensed under the MIT License.
• Uses Sentry for error telemetry (opt‑out available).
• Contributions are welcomed via the usual GitHub pull‑request workflow; detailed contribution guidelines are linked in the docs.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.