The App Store Connect intelligence layer for your coding agent. 11 tools that think, 3 slash-command workflows, and a bundled Claude Skill. Not another API wrapper.
Maintained successor to JoshuaRileyDev/app-store-connect-mcp-server (archived Feb 2026). Different angle, same API surface plus more.
npm install -g @pofky/asc-mcp
asc-mcp install-skill # one-line, optional, for auto-routed review questionsOther ASC MCP servers wrap the API and give you 80 to 293 raw endpoints. This one gives you 13 opinionated tools, 3 slash-command Prompts, and a Claude Skill that all think. Two of the tools use MCP Sampling: your own client's model does the LLM work, so there is no extra cost from this server.
| You say | What happens |
|---|---|
| "Run a preflight check on my app" | Audits metadata, character limits, screenshots, build status. Catches the issues that cause 40% of rejections. |
| "Give me a morning briefing" | Summarizes all your apps: who's in review, who got rejected, new low-rating reviews, action items. |
| "Generate release notes from my git history" | Reads commits since last tag, categorizes them, and gives you structured data to write "What's New" text. |
| "List my apps" | Shows all your iOS/macOS apps with bundle IDs |
| "Is my app in review?" | Exact review state with context ("typical time: 24-48 hours") |
| "Show me 1-star reviews" | Customer reviews filtered by rating, territory, sorted by date |
| "What were my downloads this week?" | Sales and revenue summary by territory |
No context switching. No portal. Just ask.
Step 1. Create an API key in App Store Connect > Keys (Admin or App Manager role). Download the .p8 file.
Step 2. Install:
npm install -g @pofky/asc-mcpAdd to ~/.claude/settings.json (Claude Code) or your agent's MCP config:
{
"mcpServers": {
"appstore-connect": {
"command": "asc-mcp",
"env": {
"ASC_KEY_ID": "YOUR_KEY_ID",
"ASC_ISSUER_ID": "YOUR_ISSUER_ID",
"ASC_PRIVATE_KEY_PATH": "~/.appstore/AuthKey_XXXX.p8"
}
}
}
}Step 3. Ask your agent: "List my App Store Connect apps"
Works with Claude Code, Cursor, Windsurf, Cline, and any MCP-compatible client.
| Tool | What it does |
|---|---|
list_apps |
List all your apps with name, bundle ID, SKU, platform |
app_details |
Version history, build status, release state, dates |
review_status |
Current review state with human-readable context |
| Tool | What it does | Why it matters |
|---|---|---|
list_reviews |
Customer reviews filtered by rating, territory, sort order | See what users say without opening the portal |
sales_report |
Daily/weekly/monthly downloads and revenue by territory | Know your numbers instantly |
release_preflight |
Pre-submission audit: metadata, char limits, screenshots, builds | Catches 40%+ of common rejection causes before you submit |
daily_briefing |
Morning summary across all apps: status, reviews, rejections | One call replaces 10 minutes of portal clicking |
release_notes |
Git commits since last tag, categorized for writing "What's New" | Your AI agent writes release notes from your actual changes |
keyword_insights |
Analyze keywords against iTunes search competition, difficulty ratings | See which keywords are worth targeting at a glance |
competitor_snapshot |
Look up any app: ratings, reviews, version, price, release notes | Competitive intelligence without leaving your editor |
metadata_diff |
Compare live vs pending version metadata across all locales | Verify exactly what changed before submitting |
triage_reviews |
Pulls recent reviews and clusters them into 3 to 5 themes with counts, action buckets, and quotes, using MCP Sampling | Zero extra cost: your client's model does the clustering |
draft_review_response |
Drafts a public reply to a single review in the review's locale, via Sampling. Elicits tone if your client supports it. Never auto-posts. | Apple guideline 1.2 respected; you paste into ASC yourself |
Coming next (Week 3): rejection-risk scoring against the 2026 corpus (guideline 2.3 metadata, 4.0 design, privacy-AI 5.1.2).
Get Pro | Retrieve your license key
triage_reviews and draft_review_response use the MCP Sampling primitive. When you call one of these tools, this server sends a sampling/createMessage request back to your own MCP client. Your client runs the LLM locally (Claude Desktop uses your account; Claude Code uses its session). We never call Anthropic from our side.
Outcome: review clustering and reply drafting cost you exactly what you would pay for any other Claude request, not a penny more from this server.
If your MCP client does not support Sampling yet, both tools return a structured degraded: true result with a clear explanation. Upgrade Claude Desktop or Claude Code to a recent version for full functionality.
Type these in Claude Desktop or Claude Code and the agent runs a pre-built multi-tool workflow:
| Slash command | What it does |
|---|---|
/asc-weekly-review |
Calls daily_briefing then list_reviews (low-rating, last 7 days) for every app, clusters themes, returns a single digest with 3-bullet action list. |
/asc-rejection-audit |
Given an app_id, calls release_preflight + metadata_diff + review_status, reads the result against the top 2026 rejection drivers (guideline 2.3 metadata, 4.0 design, privacy-AI 5.1.2). Produces Blocking / Likely-flagged / Safe sections. |
/asc-release-go-no-go |
Given an app_id, combines preflight + review queue + metadata diff + top competitors in the same category. Returns a direct GO or NO-GO with three supporting reasons. |
These are MCP Prompts per the spec. Zero other App Store Connect MCP ships with them.
After installing the MCP, run:
asc-mcp install-skillThis copies a small asc-review-triage Skill to ~/.claude/skills/ so Claude automatically picks up review-related questions ("any bad reviews lately?", "what do my users say?", "ratings this week?") and calls the right ASC tools without you having to explain the workflow each time.
Works on macOS, Linux, and Windows. To remove: asc-mcp uninstall-skill.
"Run a preflight check before I submit"
Release Preflight: v2.3
State: PREPARE_FOR_SUBMISSION
Platform: IOS
PASS (with 1 warning)
Warnings (recommended):
- Missing screenshot set for APP_IPHONE_67. May be required.
Passing checks: 4
- [en-US] Description OK (3874/4000 chars).
- [en-US] Keywords OK (96/100 chars).
- 2 screenshot set(s) found across 1 locale(s).
- Build 40 attached and valid.
Total: 4 pass, 1 warn, 0 fail
"Morning briefing"
Daily Briefing - 2026-04-13
2 apps in your account
Tempo: Habit Builder
- Latest: v2.3 (IOS) - Waiting for Review
- Action needed: v2.3 is Waiting for Review
- Reviews (last 3d): 5 new, avg 4.2 stars
NightOwl Weather
- Latest: v1.1 (IOS) - Live
- No new reviews in the last 3 days
"Generate release notes from git"
Git History for Release Notes
Since: v2.2.0
Commits: 8
Character limit: 4000 chars for "What's New"
New Features (3)
- feat: add habit streak calendar view
- feat: dark mode support
- add widget for home screen
Bug Fixes (2)
- fix: notification timing off by 1 hour
- fix: crash on iPad when rotating
Instructions: Write user-facing "What's New" text.
Lead with the most impactful change. Keep under 4000 chars.
| Raw API wrappers (free) | This server | |
|---|---|---|
| Tool count | 80 to 293 | 13 opinionated tools |
| MCP Prompts (slash commands) | No | Yes, 3 pre-built workflows |
| MCP Sampling (zero server-side LLM cost) | No | Yes, review triage + response drafts |
| Claude Skill bundled | No | Yes, one-line install |
| Pre-submission audit | No | Yes, catches rejections before you submit |
| Cross-app briefings | No | Yes, one call, all apps |
| Git-aware release notes | No | Yes, reads your project's commit history |
| Smart review summaries | No | Yes, theme clustering, action items |
| Setup | Build from source (Swift or Node) | npm install -g (any OS) |
| Free tier | Some | Yes, 3 tools, no account needed |
Raw wrappers give you endpoints. This gives you answers.
Your credentials never leave your machine:
- The
.p8private key is read locally. JWT tokens are generated on your computer. - API calls go directly from your machine to
api.appstoreconnect.apple.com. - The license server sees only your license key string. Zero Apple data, zero credentials.
- Fully open source. Read the code.
- Claude Code (Anthropic)
- Cursor
- Windsurf
- Cline
- Any client supporting the Model Context Protocol
- Node.js 18+
- Apple Developer Program membership
- App Store Connect API key (Admin or App Manager role)
This project is not affiliated with, endorsed by, or sponsored by Apple Inc. Apple, App Store, App Store Connect, TestFlight, iOS, and macOS are trademarks of Apple Inc.
MIT