DOCUMENTATION

Connect ChatGPT to Xcode.

Authorize a project locally. Connect your client. Start with one verifiable read-only request.

On this page · jump to a step

To let ChatGPT work with local Xcode, you need the Relayard Mac app, an authorized project and a client supporting remote MCP / OAuth and the required tools.The hosted relay connects the client to native tools on your Mac. Verify reading before allowing edits and execution.

Who is it for, and what does it solve?

Developers maintaining an iOS or macOS project who want AI to read current source and use local tools. The project does not have to move to a new cloud build environment. You still review the target project and capabilities.

Connection and data flow

The Mac app connects outward to the hosted relay. ChatGPT authorizes through OAuth at the project's HTTPS MCP address, then requests approved operations. Local results return through the relay. Possessing a URL does not grant access; requested source, logs and images may leave your Mac.

Installation and availability

First obtain a working Relayard Mac app build, install it according to its release instructions and open the same project in Xcode. System, chip and version requirements depend on the actual release.

When is this a poor fit?

This workflow cannot proceed without Mac / Xcode, the account's MCP connection capability or model support for the tools. It also does not fit a requirement that even requested source excerpts must remain entirely offline.

Check current availability · See the complete small verification task

Before you start

  • A working Relayard build on your Mac and a full Xcode installation that exposes native MCP capabilities.
  • A project that opens correctly in Xcode, with editor changes saved.
  • An AI client with remote MCP and OAuth support, enabled for your account or workspace.
Start with a sample project before authorizing your day-to-day work. Use the system requirements published with the specific build you receive.

1. Authorize a project on your Mac

Open Relayard, sign in to your product account and register your Mac. Select the actual Xcode.app, source directory and corresponding .xcodeproj or .xcworkspace. Open that same project in Xcode.

Enable external tool access in Xcode’s Intelligence / Model Context Protocol settings and complete the local authorization prompt. Setting names can vary by Xcode version.

2. Discover and review tools

Connect to Xcode from Relayard’s project and tools view. Check the project path and binding, then review the native tools that were actually discovered. Start with read-only access.

Editing, execution and Simulator actions are separate capabilities. Builds, previews and tests can execute project code; they are not ordinary file reads.

3. Create and copy your MCP URL

Create an endpoint for the selected project and resume the connection. Check native readiness, the service connection and catalog confirmation separately. Copy the complete HTTPS MCP URL shown in the app.

The endpoint URL is not a password. Access still requires OAuth. Keep access tokens, verification codes and sensitive diagnostic content out of public pages.

4. Connect from ChatGPT

Use the App, connector or MCP connection entry available to your account to add the URL. Choose OAuth, sign in with the same Relayard account, and review the target project and permissions on the consent screen.

The client interface depends on your account, workspace policy and product version. If the entry is missing, confirm client support first. After authorization, inspect the client’s actual imported tools; a tool-search result count alone does not establish the full catalog.

5. Make a read-only request

Use this prompt to verify the project and file-reading path. Obtain a real result before moving on to edits, builds or previews.

TRY THIS IN CHATGPT
Use Relayard to check the current project and connection status.
List the tools actually available to this client.
Read one source file from the selected project, without modifying it.
Report the project, the file read, and any permission or readiness issue.
Do not create a new binding, run a build or change code.

After your first connection

Continue with builds and previews. For missing tools or an unexpected state, follow the troubleshooting guide before creating another endpoint or rebinding the project.

Continue from Codex

Select a Pro model in Chat mode and connect your local project through Relayard. Pro here refers to the Chat model or thinking option, with an allowance separate from Work and Codex. Check that you have Chat Pro usage remaining and that the selected model supports the required tools.

  1. Save the handoff state.Save editor files, preserve a Git checkpoint, and record completed work, remaining tasks and the latest build errors. Avoid two agents editing the project at once.
  2. Bring a short handoff.Give ChatGPT the goal and current progress. Relayard connects the project; it does not automatically import your Codex conversation.
  3. Continue from the actual files.Have the model inspect the binding and relevant source, work through the scoped release requirements, then check real build or preview results. One request can cover multiple steps; complex projects or blockers may still need follow-up messages.
TRY THIS IN CHATGPT
Use Relayard to work through a complete release iteration for this Xcode project.
Release goal and user scenario: fill in here.
Features and fixes in scope: fill in here.
Completed work and existing changes: fill in here.
Out-of-scope areas: fill in here.
Acceptance criteria: fill in here.

First confirm the binding, permissions and available tools, then inspect the relevant source.
Make a short implementation plan and work through it; do not stop after planning.
Preserve existing changes and use authorized builds, tests or previews to check each stage.
If blocked, preserve the error and explain what is needed. Do not invent passing results.
Finish with changes, verification evidence, remaining items and acceptance steps.
Do not assume you inherited the Codex conversation. Do not automatically commit or publish.
Usage note: Pro models in Chat use an allowance separate from Work and Codex. Counts and reset periods depend on the plan and model; check your account. A release iteration is a workflow goal, not a promise that it always consumes one message or finishes in one request.
Read OpenAI’s usage documentation

Documentation updated: 2026-10-08