ASO MCP server for Claude: App Store and Google Play data

What can I ask an ASO MCP server?

Anything the measurement already holds for the apps and storefronts you track. The assistant picks the tool; the table shows which one answers which kind of question.

For a full review of one app in one pass, the free ASO audit skill tells Claude which of these tools to call, in which order, and how to report the result.

Setting it up

Add this to your MCP client's config — claude_desktop_config.json for Claude Desktop. The API key comes from your Storelift settings.

{
  "mcpServers": {
    "storelift": {
      "command": "npx",
      "args": ["-y", "storelift-mcp"],
      "env": { "STORELIFT_API_KEY": "sl_live_..." }
    }
  }
}

In Claude Code it is one line:

claude mcp add storelift -e STORELIFT_API_KEY=sl_live_... -- npx -y storelift-mcp

There is also a hosted endpoint, with nothing to install: https://storelift.net/mcp over Streamable HTTP. It serves the same tools and works from CI or a cloud agent.

Adding it to Claude.ai as a connector

The hosted endpoint supports MCP sign-in (OAuth), so there is no key to copy. In Claude.ai, open Customize → Connectors, click +, choose Add custom connector and paste https://storelift.net/mcp (on Team and Enterprise an owner adds it under Organization settings → Connectors, then members click Connect). Claude sends you to a Storelift page that names the app asking and the address you will return to; sign in if needed and choose Allow.

The connection reads the same data as an API key, and every change it proposes stays a preview until you approve it. It appears in Storelift under Settings → API → Connected apps, where removing it cuts access on the next request.

Claude Code signs in the same way. Add the endpoint without a header, then run /mcp inside Claude Code to authenticate:

claude mcp add --transport http storelift https://storelift.net/mcp

Clients that let you set a header can use an API key instead:

claude mcp add --transport http storelift https://storelift.net/mcp --header "Authorization: Bearer sl_live_..."

Step-by-step guides per client

Each guide covers the version, requirements, setup, a first read, removing access and the error replies you may see.

• Claude
• Claude Code
• Cursor

Once connected, three first tasks show what to ask and the answer to expect: rank change over a week, who is ahead on the keywords that fell, and a keyword with no measurement today.

How the local server works

The server is a single file and talks JSON-RPC over stdio with no dependencies, so npx fetches it on first run and there is nothing to keep updated. It is registered in the MCP registry as net.storelift/storelift.

What the 85 tools do

Every tool, with what it returns, is in the tool directory. Read tools read the same measurement the dashboard shows; write tools return a preview first.

Questions to try first

Each of these is answered by the tools above, from your own tracked data. The tool names are there so you can see what the assistant calls.

• “Which of my US keywords lost ground this week, and who is above me on them now?” — get_keywords, get_history, get_rivals
• “Which apps rank above me on my main keyword, and what do their names have in common?” — get_rivals, get_store_page
• “Summarise my recent reviews that mention a crash or a bug.” — get_reviews
• “Where does my app sit in its category chart in Turkey?” — get_charts
• “Do assistants name my app when asked about my category?” — get_ai_visibility, run_ai_visibility_check, which reads dated samples of AI assistant answers, not what every user sees; these MCP tools require the ai:read scope. A new run uses the workspace's available AI visibility allowance.
• “Walk me through this week's ASO review: what should I work on first, which keywords moved, and who is ahead of me on them?” — get_weekly_plan, get_keywords, get_rivals

The server also publishes these 6 questions as MCP prompts, so a client that lists prompts offers them ready to run.

Why an MCP server rather than a plain API

There is a plain REST API too, and for a script that is the right thing. MCP earns its place when the reader is an assistant rather than a program: the tools carry their own descriptions, so the assistant picks the right call and the right parameters without you writing the integration.

The practical difference is what you can ask. “Which keywords did I lose ground on since the last release, and who took the position” is three calls and a join in a script; in an MCP client it is one sentence.

What the MCP server writes

20 tools change something, and each runs in two steps: the first call returns a preview of what would change and a confirm token, and nothing is written until the assistant sends the token back. An API key writes only when it carries a write scope.

• add_app, add_keywords, remove_keywords, add_rivals, remove_rivals — your tracking setup. A term added today gets its first rank in the next measurement.
• reply_review — a developer reply on an App Store review, through your App Store Connect key.
• update_listing — the listing text of an editable App Store version, through your App Store Connect key.
• pause_campaign, resume_campaign, set_daily_budget — Apple Ads campaigns; one budget step moves at most 50%, within a daily action allowance.
• create_share_link — a public view link to one storefront's dashboard.
• export_data — one app's keywords, rank history or reviews as a CSV or JSON file, through a download link valid for 7 days.
• mark_alerts_read, create_alert_rule, delete_alert_rule — the alert feed and the rules behind it: which change alerts you, and on which channel.

Read next