--- title: Is there an MCP server for Xcode? Limrun's answer description: Apple's Xcode MCP server needs a Mac running Xcode, so here is how Linux-based coding agents get Xcode builds and cloud simulators from Limrun instead. author: Limrun date: 2026-10-26 url: https://limrun.com/blog/xcode-mcp-server/ --- # Is there an MCP server for Xcode? Limrun's answer Yes. Since Xcode 26.3, Apple ships an MCP server for Xcode ([`xcrun mcpbridge`](https://developer.apple.com/documentation/xcode/giving-external-agents-access-to-xcode)) that local agents such as Claude Code and Codex use to edit, build, run tests, and render previews, and Xcode 27, released in September 2026, adds simulator interaction through Device Hub. All of it needs Xcode running on a Mac with the project open. For Linux-based coding agents in sandboxes or cloud runners, Limrun's remote MCP server (`https://mcp.limrun.com/mcp`) gives them cloud iOS simulators and Android emulators without a local Mac, the `lim` CLI runs Xcode builds on Limrun's Macs, and the Limrun skills, which the MCP server also hands out through `get-agent-skill`, teach the agent the CLI. Platform teams can instead attach Limrun to the Linux sandbox their agents already use. ## What does Apple's Xcode MCP support cover? Apple's Xcode 26.3 MCP support lets a Mac-based assistant read and edit Xcode project files, search documentation, build, run tests, grab previews, and change project settings, and [Xcode 27](https://www.apple.com/newsroom/2026/06/apple-aids-app-development-with-new-intelligence-frameworks-and-advanced-tools/) adds simulator interaction through Device Hub. It needs Xcode running on a Mac with the project open, so it does not turn a Linux sandbox into a Mac build environment or expose simulator control to a Linux agent. [Apple's Xcode 26.3 announcement](https://www.apple.com/newsroom/2026/02/xcode-26-point-3-unlocks-the-power-of-agentic-coding) introduced the Model Context Protocol integration with Claude Agent and OpenAI Codex. [Ars Technica's coverage of Xcode 26.3](https://arstechnica.com/apple/2026/02/xcode-26-3-adds-support-for-claude-codex-and-other-agentic-tools-via-mcp) reported that the machine-invocable interfaces cover project files, documentation search, and project settings. [Apple's WWDC26 Platforms State of the Union](https://developer.apple.com/videos/play/wwdc2026/102) adds that Xcode also connects MCP-powered tools to external sources such as Figma and GitHub. Those integrations make Xcode an agentic IDE on macOS. A Linux runner still cannot retrieve an Apple signing identity, run an Xcode build, or tap through an iOS simulator. In practice, teams with Linux agents should treat Xcode's native MCP support as an IDE-level feature, not a remote build capability. ## Is there an MCP server for Xcode? Yes. Apple's Xcode ships one for agents on the same Mac, and Limrun runs a remote MCP server with 16 tools that create, drive, stream, and delete iOS simulators and Android emulators from a Linux sandbox. Teams do not need to move their agent platform. This is the practical answer when Apple's built-in MCP support stays on macOS. The server exposes operations for creating device instances with apps you already uploaded preinstalled, deleting them, streaming the live screen, and sending taps or reading screenshots on a simulator or emulator. Builds, source sync, and build logs stay on the `lim` CLI. Engineers can point Claude Code, Codex CLI, Gemini CLI, or Cursor at it by installing the Limrun plugin, and Claude and ChatGPT users add it from the Claude connectors directory or the ChatGPT plugin, as the [mobile simulator MCP guide](https://docs.limrun.com/docs/guides/mobile-simulator-mcp) shows. The [remote MCP server docs](https://docs.limrun.com/docs/agents/remote-mcp) show the endpoint and tool list, so a team can wire the server into an existing sandbox in a short session. A single device layer then serves several agents without separate integrations. Because the server and the builds run remotely, the agent never needs Xcode, a signing store, or a simulator runtime installed locally. ## Is there a Claude skill for Xcode? Yes. Limrun ships Agent Skills for Claude, Cursor, and Codex through its lim CLI, including a Claude skill for Xcode. A Claude agent can start an Xcode build, install the app to a simulator, and drive device actions from a Linux sandbox. The skill gives Claude a portable Xcode capability. Instead of prompting a human to open a Mac, the agent calls the skill to create an instance with the `lim` CLI. The [Claude Code web guide](https://docs.limrun.com/docs/agents/claude-code-web) documents how a web-based Claude agent follows that path, so teams can reproduce it in their own setup. Underneath, the skill teaches the agent the `lim` CLI commands that build the app and attach a simulator, and the MCP server hands out the same skill through `get-agent-skill`. Because the skill set also covers Cursor and Codex, the same infrastructure works for mixed agent fleets. ## How should remote Xcode builds run from a Linux sandbox? A Claude skill only helps when the remote Xcode builds behind it run cleanly on Mac hardware. The build should leave the Linux checkout and agent edits in place, then hand only Apple work to the Mac fleet. Five pieces make that practical: changed-file sync, live logs, a warm cache, signing, and a TestFlight or App Store handoff. Limrun's Xcode build sandbox runs on real Macs and syncs only changed source files. It streams build logs as they arrive and uses an S3-backed cache for warm builds. For platform teams, those details matter because agents often trigger builds frequently. A full re-upload on every prompt would slow the loop. Changed-file sync keeps the agent's edit-build-test cycle short, while live logs let a human inspect a failing build without switching tools. Warm builds matter when the agent iterates on a small change several times in one session. The [remote Xcode build guide](https://docs.limrun.com/docs/ios/build-with-xcode) shows how to run that path from a Linux pipeline. After a successful compile, the platform signs with a manual profile or App Store Connect API keys, then returns a signed IPA or pushes the build to TestFlight and App Store Connect. These build operations run through the same `lim xcode build` command, rather than a separate CI product. That release handoff is what turns a coding agent from a build helper into a delivery path. ## What should a cloud Xcode for agents do after the build? Once a build returns a signed app, cloud Xcode for agents should install it to a simulator, let the agent drive the UI, share a preview link, and run mobile tests. Otherwise the pipeline stops at compilation and never confirms the app behaves. The platform covers those four stages on the same control plane. The simulator fleet covers iPhone, iPad, and Apple Watch. Through MCP and the `lim` CLI, an agent can send multi-step gestures, capture screenshots, inspect the accessibility hierarchy, inject camera feeds, manage apps, read logs, and record video. That control is what makes an MCP server for Xcode useful after compilation: the agent checks behavior instead of only reading a build log. Batched actions run next to the simulator with no round trip between steps, so an agent can reproduce a multi-step UI check without a human clicking through a browser. While the simulator runs, the agent can share its signed stream link, and anyone who has it can watch and control the app in a browser without a Limrun account until the instance stops. On a pull request, a [preview link](https://docs.limrun.com/docs/ci/pr-previews) opens the uploaded build in a fresh simulator for members of your Limrun organization, so a reviewer can click it and try the same build the agent verified. For real phones, `lim xcode build ... --upload-to-appstore` sends a signed build to App Store Connect, and testers install it from [TestFlight](https://docs.limrun.com/docs/ios/sign-and-distribute) on their own phones. That moves the output from a CI log to a demo, a manual QA pass, or a stakeholder review. Sign up free in the [Limrun console](https://console.limrun.com), where new accounts get trial credits and you create your API key, then follow the [quickstart](https://docs.limrun.com/docs/quickstart). Plans are on [limrun.com/pricing](https://limrun.com/pricing). From there, the same cloud Xcode for agents attaches to GitLab, CircleCI, Buildkite, or Jenkins. ## FAQ ### Does Xcode's built-in MCP support run in a Linux container? No. Xcode's built-in MCP support runs inside Xcode on macOS and targets a local Apple environment. A Linux container has no Xcode, signing identity, or simulator runtime to act on. Limrun gives the Linux agent those Mac-backed capabilities through a remote MCP server for cloud simulators and the `lim` CLI for Xcode builds and signing. ### Can I use the same MCP server with Cursor or Codex? Yes. The remote endpoint accepts connections from Claude Code, Claude Desktop, Cursor, and Codex, so teams can run mixed agents without separate integrations. In Cursor and Codex, the Limrun plugin installs the server together with the Limrun skills, and `lim skills install` adds the same skills from the CLI. ### How does an iOS simulator MCP control the device? The iOS simulator MCP exposes batched taps, typing, and gestures, screenshots with the accessibility element tree, listing, launching, and stopping apps, app logs, and screen recording. Batched actions run next to the simulator with no round trip between steps. Apps are preinstalled when the agent creates the simulator, and installing a fresh build or playing a video as the camera feed (`lim ios camera play`) goes through the `lim` CLI. An agent calls those tools to drive a UI flow and read back the UI state after each step. ### What Apple platforms can a remote Xcode build target? Remote Xcode builds support Swift and React Native and can target iOS, watchOS, tvOS, and visionOS. The Mac fleet handles Swift compilation, asset catalogs, and code signing, while the Linux sandbox keeps the checkout and agent edits. ### Can an agent install a build to a real phone? Yes, through TestFlight. An agent can sign the build and send it to App Store Connect with `lim xcode build ... --upload-to-appstore`, and testers install it from TestFlight on their own phones. That path is useful for a manual QA pass or a stakeholder demo after the agent has already verified the flow in the simulator. The device gets the signed app without the agent needing a local Mac. ### Which mobile test frameworks work with Limrun's iOS simulators? Limrun's iOS simulators run XCTest suites from CI and work with Appium, Maestro, and Detox on iOS. That covers native test bundles and agent-driven UI checks. On Android, the same control plane supports Android 14 and 15 emulators through the lim CLI, TypeScript SDK, MCP, or an ADB tunnel. ### Does the platform expose its API key to the browser? No. The API key stays on your backend or in your CLI config. That boundary matters for team security, because a shared preview link or a live simulator embed should not hand credentials to the user. The browser only gets a token for one running instance (for an embed) or that instance's signed stream link, which works only while the instance runs and lets anyone holding it use that one device.