xcode-cloud MCPサーバー

xcode-cloud-mcp

MCP server for discovering Xcode Cloud products, inspecting and editing workflows, starting and monitoring build runs, and retrieving build issues, logs, test summaries, and UI test artifacts through the App Store Connect API.

Features

Feature	Tool(s)	Example use	Example return
Discover products	list_products	"Show me the Xcode Cloud products available in this account."	Demo App, productType: APP, createdDate: 2026-03-30T10:00:00Z
Discover workflows	list_workflows	"List the workflows for product def456."	Feature Branch, description, isEnabled: true, containerFilePath: Chauffeur.xcodeproj
Inspect workflow configuration	get_workflow_details	"Show me the full workflow details for abc123, including environment and actions."	general, environment, startConditions, actions, postActions
Monitor running or recent builds	list_build_runs	"Show me the running builds for workflow abc123 so I can monitor them."	number: 93, executionProgress: RUNNING, completionStatus: null, startedDate: ...
Start an Xcode Cloud build	start_build	"Start one build for workflow abc123."	newly created build-run ID, number, and execution state
Enable or disable a workflow	set_workflow_enabled	"Disable workflow abc123 while we are testing new settings."	operation.type: set_workflow_enabled, workflow.general.isEnabled: false
Update name, description, or clean mode	update_workflow_general	"Rename workflow abc123 to Feature Branch v2 and adjust its description."	changedFields: [name, description], updated workflow.general
Update start conditions explicitly	update_workflow_start_conditions	"Change workflow abc123 so pull-request builds no longer auto-cancel."	updated workflow.startConditions.pullRequest.autoCancel: false
Replace the workflow action list	update_workflow_actions	"Remove the archive action from workflow abc123, then add it back once the experiment is done."	actionCount: 4 after removal, then actionCount: 5 after restore
See build health quickly	get_build_issues	"What went wrong in the latest failing build for workflow abc123?"	issueCounts: { errors: 1, testFailures: 3, warnings: 2 }
Read compact build log summaries	get_build_logs	"Retrieve logs of build 81 and summarize the failure."	failedTests, highlights, excerpt, savedLogsDirectory
Materialize logs for local grep	materialize_build_logs	"Download the logs for build 81 so I can grep them locally."	savedLogsDirectory: /var/folders/..., savedLogs: [...]
Summarize test outcomes	get_test_results	"Summarize the test results for the latest failing build."	testFailures, issueCounts, summary
Jump straight to failed tests	get_failed_tests	"What tests failed in build 81?"	displayExpiryDateReturnsFormattedDateWhenExpiryDateExists(), assertion message, saved log paths
Retrieve UI test artifacts	get_test_artifacts	"Show me the screenshots and videos from the latest failing UI test run."	screenshots, videos, resultBundles, downloadUrl
Clean up local temp files	cleanup_saved_logs	"Remove saved logs older than 24 hours."	removedDirectories: [...], retainedDirectories: [...]

Build lookup is workflow-scoped. Retrieval tools accept a direct buildRunId, or a workflowId plus buildNumber, or a workflowId plus buildSelector: "latest" | "latestFailing".

list_products and list_workflows automatically paginate through all results.

list_build_runs supports status: "all" | "failed" | "succeeded" | "running" | "pending" and an optional limit, which defaults to 20, so agents can poll active workflows without post-processing every run locally or inflating MCP response size.

Build Start Behavior

start_build starts exactly one build run through Apple's public POST /v1/ciBuildRuns endpoint. workflowId is required. The optional sourceBranchOrTagId and pullRequestId values are App Store Connect resource IDs, not branch names, tag names, or pull-request numbers; buildRunId may be a bare resource ID or an xcode-cloud://build-run/... URI. The optional clean value applies only to the new run and does not modify the workflow.

The operation is externally side-effecting and non-idempotent: repeating the same invocation can create another build. The server sends one POST and does not automatically retry an ambiguous failure; inspect list_build_runs before deciding whether to try again.

Apple's public App Store Connect API does not expose an operation for canceling an individual Xcode Cloud build run. This server therefore provides no cancellation tool and does not disable or otherwise mutate a workflow as a cancellation surrogate.

Requirements

• Node.js 20+
• App Store Connect API credentials with access to Xcode Cloud

Environment Variables

Primary names:

• APPSTORE_CONNECT_API_KEY_ID
• APPSTORE_CONNECT_API_ISSUER_ID
• APPSTORE_CONNECT_API_KEY_CONTENT

Compatibility aliases:

• APP_STORE_KEY_ID
• APP_STORE_ISSUER_ID
• APP_STORE_PRIVATE_KEY

The private key can be passed as literal multi-line PEM content or as a string with escaped \n.

Claude Setup

claude mcp add xcode-cloud \
  --env APPSTORE_CONNECT_API_KEY_ID="$APPSTORE_CONNECT_API_KEY_ID" \
  --env APPSTORE_CONNECT_API_ISSUER_ID="$APPSTORE_CONNECT_API_ISSUER_ID" \
  --env APPSTORE_CONNECT_API_KEY_CONTENT="$APPSTORE_CONNECT_API_KEY_CONTENT" \
  -- npx -y @thatfactory/xcode-cloud-mcp

Codex Setup

codex mcp add xcode-cloud \
  --env APPSTORE_CONNECT_API_KEY_ID="$APPSTORE_CONNECT_API_KEY_ID" \
  --env APPSTORE_CONNECT_API_ISSUER_ID="$APPSTORE_CONNECT_API_ISSUER_ID" \
  --env APPSTORE_CONNECT_API_KEY_CONTENT="$APPSTORE_CONNECT_API_KEY_CONTENT" \
  -- npx -y @thatfactory/xcode-cloud-mcp

Available Tools

• list_products()
• list_workflows(productId)
• get_workflow_details(workflowId)
• list_build_runs(workflowId, limit?, status?)
• start_build(workflowId, clean?, sourceBranchOrTagId?, pullRequestId?, buildRunId?)
• set_workflow_enabled(workflowId, enabled)
• update_workflow_general(workflowId, name?, description?, clean?)
• update_workflow_start_conditions(workflowId, branchStartCondition?, manualBranchStartCondition?, pullRequestStartCondition?, manualPullRequestStartCondition?, scheduledStartCondition?, tagStartCondition?, manualTagStartCondition?)
• update_workflow_actions(workflowId, actions)
• configure_manual_release_candidate(workflowId, scheme, branch, buildDistributionAudience)
• get_build_issues(buildRunId? workflowId? buildNumber? buildSelector?)
• get_build_logs(buildRunId? workflowId? buildNumber? buildSelector?, maxCharacters?)
• materialize_build_logs(buildRunId? workflowId? buildNumber? buildSelector?)
• get_test_results(buildRunId? workflowId? buildNumber? buildSelector?)
• get_failed_tests(buildRunId? workflowId? buildNumber? buildSelector?)
• get_test_artifacts(buildRunId? workflowId? buildNumber? buildSelector?)
• cleanup_saved_logs(buildRunId?, maxAgeHours?)

Log Retrieval Behavior

get_build_logs keeps the MCP response compact on purpose:

• it downloads and extracts text-like build log artifacts to a temporary local directory
• it returns savedLogsDirectory and savedLogs so local agents can inspect the extracted files with rg, grep, or cat
• it returns a compact failedTests summary, highlights, and a capped excerpt
• even if a caller passes a very large maxCharacters, the inline excerpt is clamped to avoid oversized MCP responses

Recommended agent workflow:

1. Call get_failed_tests or get_build_logs.
2. Read savedLogsDirectory.
3. Use rg inside that directory to inspect the exact failing test or assertion.
4. If needed, call cleanup_saved_logs when the investigation is done.

Temporary logs are written under the system temp directory in a path like:

/tmp/xcode-cloud-mcp/build-logs/<buildRunId>

On macOS this typically resolves to a path under /var/folders/.../T/.

Cleanup policy:

• each new call for the same buildRunId deletes and recreates that build-specific temp directory first
• older build directories are pruned automatically when they are older than 24 hours
• you can also call cleanup_saved_logs directly for one buildRunId or for all directories older than a chosen retention window

Example Prompts

Retrieve logs of the latest failing build for workflow abc123.

Retrieve logs of build 81, then inspect the returned savedLogsDirectory and grep for Expectation failed.

Get the failed tests for build 81, then open the saved logs directory and inspect the failing test in context.

Retrieve logs of build number 42 for workflow abc123.

Show me the latest failing UI test artifacts for workflow abc123.

List the workflows for product def456 and then summarize the latest build.

Show me the full workflow details for workflow abc123, including environment, start conditions, actions, and whether it is enabled.

Disable workflow abc123, remove the archive action, then restore the original action list after the experiment.

Workflow Details Behavior

get_workflow_details returns the live workflow configuration exposed by App Store Connect, grouped into:

• general
• environment
• startConditions
• actions
• postActions

Notes:

• environment includes repository, xcodeVersion, and macOsVersion when App Store Connect returns them.
• actions includes action type, scheme, platform, destination, required-to-pass state, and test-plan details when present.
• postActions is currently returned as an empty array with a note because Apple does not expose workflow post-actions in its public API. This does not mean no post-actions are configured; testFlightDistribution reports UNSUPPORTED_BY_APPLE_API and an actionable next step.

Workflow Update Behavior

The workflow update tools are intentionally explicit:

• set_workflow_enabled only toggles isEnabled
• update_workflow_general only changes name, description, and clean
• update_workflow_start_conditions only changes the start-condition objects you pass
• update_workflow_actions replaces the full actions array, so callers should fetch the current workflow first and then send the final desired action list

Important restriction:

• if the workflow has Restrict Editing enabled in Xcode Cloud, edits can fail even when the App Store Connect API key has App Manager access
• for MCP edits to work reliably, disable the Restrict Editing checkbox for that workflow before using the write tools
• if Apple still rejects the request after that, use a stronger API key role such as Admin

Local Development

Install dependencies:

npm install

Run tests:

npm test

Build the package:

npm run build

Manual TestFlight release candidates

configure_manual_release_candidate converts an existing workflow to one macOS archive action (ARCHIVE, MACOS, ANY_MAC) with manual starts from an exact branch. It replaces the entire action list and all seven start conditions in one PATCH, clearing automatic branch, pull-request, tag, and scheduled starts plus manual tag and pull-request starts. Fetch get_workflow_details first if you need to retain the original configuration. The preset preserves the workflow's enabled state, name, environment, and clean-build setting; enable a disabled workflow separately when ready.

{
  "workflowId": "abc123",
  "scheme": "Headroom",
  "branch": "main",
  "buildDistributionAudience": "APP_STORE_ELIGIBLE"
}

The audience is required and cannot be null in this preset:

• APP_STORE_ELIGIBLE: Deployment Preparation = TestFlight and App Store.
• INTERNAL_ONLY: Deployment Preparation = TestFlight internal testing only.

The general update_workflow_actions tool accepts only these two audience values or null/omission (Deployment Preparation = None). Use the preset when requesting a TestFlight release candidate: null/omission is rejected before any API mutation. Workflow details retain the raw buildDistributionAudience and add the human-readable deploymentPreparation value for each action.

Creating an archive, making it eligible for TestFlight, and assigning its processed build to tester groups are separate steps. This preset configures eligibility; it does not itself start a build, guarantee successful upload/processing, or assign testers. Call start_build explicitly to create a build run. To distribute automatically, edit the workflow in Xcode or App Store Connect, add a TestFlight Internal Testing post-action, and select the internal group.

Apple's OpenAPI specification, version 4.4.1, inspected on 2026-09-07, exposes no post-action field or relationship in CiWorkflow, CiWorkflowCreateRequest, or CiWorkflowUpdateRequest, and no workflow post-action endpoint. Consequently, workflow responses explicitly report testFlightDistribution.status: "UNSUPPORTED_BY_APPLE_API" and group assignment as UNKNOWN; the legacy empty postActions array is not evidence that no post-actions exist. See Apple's BuildAudienceType and TestFlight distribution guide.

Waiting for completion, TestFlight processing, beta-group assignment, and broader release orchestration remain outside this server's current scope and would not be native Xcode Cloud post-actions.