Guide
AI ASO With Claude Code: A Weekly Routine Using Real Data
AI ASO works when your agent reads real ranks instead of guessing. Connect Claude Code to your data over MCP and run a 30-minute weekly review, change and log.
AI ASO means using an AI agent to do the thinking part of App Store optimization: reading your ranks, spotting patterns, and drafting keyword fields and subtitles. It only works when the agent reads real data instead of guessing, so connect Claude Code (or Codex, OpenCode, or a local Ollama model) to your rank history and listing through an MCP server or plain files, then run the same 30-minute routine every week.
This guide gives you the routine, the prompts, and the guardrails. AppMeerkat’s own agent integration ships with the app, which is launching soon; until then, the file-based version below works with any tracker.
Why the Data Matters More Than the Model
Ask any chatbot “What should my App Store keywords be?” and you’ll get a confident answer with invented search volumes. The model has never seen your ranks, today’s top 10, or Apple’s Popularity for your country.
The fix is giving the agent tools. The Model Context Protocol (MCP) is an open standard for that: a server exposes tools like “get rank history” or “search the store”, and the AI client calls them. Claude Code supports MCP servers directly; its MCP documentation shows the setup:
# A local server that runs as a process on your machine
claude mcp add --scope user <name> -- <command> [args...]
# Check it inside Claude Code
/mcp
With --scope user, the server is available in every project. Run claude mcp list to see each server’s status.
Option A: AppMeerkat’s MCP Server (Ships With the App)
AppMeerkat includes a local MCP server inside the desktop app. Settings › AI has a Connect button for Claude Code, Codex, OpenCode, Cursor and Claude Desktop; it shows the exact config it will write, writes it after you confirm, and checks that it works. The tools include rank history, the country board, scores, competitor keywords, keyword gaps, live store search, and listing drafts. See ASO MCP server for the full list.
Two rules are built in:
- Every change needs your approval. Read tools run freely. Anything that changes data, like adding keywords or creating a draft, opens an approval dialog in AppMeerkat with the exact change.
- Drafts only. Listing changes are saved as drafts in App Store Connect or Google Play Console. Nothing is submitted; you press publish.
Option B: Files in a Folder (Works Today)
No MCP server yet? Make a folder Claude Code can read:
aso/
listing-en-US.md # current name, subtitle, keyword field, description
ranks-2026-11-21.csv # export from your tracker: keyword, country, date, rank
changes.md # what you changed, and when
The routine is the same; the agent reads the CSV instead of calling a tool. It’s slower to keep current, and the agent can’t search the store live, but it stops the guessing.
The Weekly Routine (About 30 Minutes)
Pick a fixed day. Monday works well because it compares a full week.
1. Review (5 minutes)
“Compare my ranks for the last 7 days with the 7 days before, per country. List the 10 biggest gains and 10 biggest drops, using weekly medians. Ignore days marked Not Checked. Don’t state any number you didn’t read from the data.”
Medians matter: single days are noisy, and a failed check isn’t a drop. Our guide on tracking keyword rankings in every country explains why.
2. Investigate One Drop (10 minutes)
Pick the drop that matters most, not the biggest number.
“My rank for ‘bill reminder’ in the UK fell from a median of 8 to 19. Get the current top 10 and compare their names and subtitles with mine. What changed?”
Typical findings: a new competitor with the exact phrase in its name, or your own last update removed a word.
3. Draft One Change (10 minutes)
“Write three subtitle options for my English (U.K.) listing. Each must be 30 characters or fewer. Don’t repeat any word from my app name. No competitor names, no claims like ‘best’. Show the character count for each.”
Then check the counts yourself. Models miscount characters, especially with symbols like “&”. Paste the winner into the free title and subtitle preview or, for keyword fields, the keyword field optimizer.
4. Approve, Draft, Publish (3 minutes)
With AppMeerkat, approve the draft and it’s saved to App Store Connect or Play Console. Without it, copy the text in yourself. Either way, you publish. On the App Store, subtitle and keyword changes go out with your next app version, so you may batch two weeks of changes into one release.
5. Log It (2 minutes)
“Add a note: ‘UK subtitle changed to …’ on the release date.”
Next Monday’s review starts from this note.
Prompts That Work Well
| Job | Prompt |
|---|---|
| Find gaps | “Which keywords do my top three competitors rank in the top 10 for in the US, where I’m outside the top 50?” |
| Localize | “Propose a Spanish (Mexico) keyword field for my app. 100 characters, commas, no spaces, no plurals. Explain each word in English.” |
| Explain drops | “Which keywords dropped in three or more countries this week? Is it one change or several?” |
| Clean a keyword field | “Find words in my keyword field that repeat my name, subtitle or category, or are plurals of other words.” |
For the rules behind the keyword prompts, see our keyword field guide.
Guardrails
- No numbers without a source. Put “don’t state any number you didn’t read from the data” in your prompt. AppMeerkat’s built-in instructions say the same.
- One change at a time. If you change the name, subtitle and keywords in one release, you can’t tell which one worked.
- You check the limits. Apple’s name and subtitle are 30 characters each and the keyword field is 100 (Apple: Product page); Play’s short description is 80 (Google Play best practices). Verify before you publish.
- Native speakers for localization. The agent’s first draft is a start. Have someone who speaks the language read it.
- Keep competitor names out. Apple forbids “competing app names” in keywords; the agent will sometimes suggest them.
Limits
- The agent’s judgment can be wrong. It’s good at patterns and drafts, less good at knowing your users. That’s why every change waits for your approval.
- Local models vary. With Ollama, choose a model that supports tool calling; small models struggle with long, multi-step reviews.
- You use your own AI. AppMeerkat doesn’t sell AI usage, so your client’s own limits and terms apply.
- Data still has gaps. Apple’s Popularity floors at 5 for many keywords, and every other demand number is an estimate. See Apple search popularity.
AppMeerkat stands watch over your ranks every day so the weekly review has something real to read. Get notified at launch.
Questions
Can Claude Code do ASO on its own?
It can write and reason well, but it doesn't know your ranks, the top 10 for your keywords or Apple's Popularity. Without real data it will guess numbers. Give it data through files or an MCP server first.
What is an ASO MCP server?
A program that gives an AI client tools to read and change ASO data, such as rank history and keyword lists, using the Model Context Protocol. Claude Code, Codex and other clients can connect to one.
Does AppMeerkat's MCP server work today?
It ships inside the AppMeerkat desktop app, which is launching soon. Until then, you can run the same routine with CSV files from any rank tracker in a folder Claude Code can read.
Can the AI publish my App Store listing?
Not through AppMeerkat. Its tools save drafts to App Store Connect and Google Play Console, and you press publish yourself. There's no tool that submits for review.
Do I need an API key to use Claude Code with AppMeerkat?
No. AppMeerkat runs the Claude Code you already have, signed in with your own account. It never sees your AI credentials. Codex, OpenCode and Ollama work the same way.
Read next
Launching soon
Let a meerkat watch your ranks.
AppMeerkat tracks every keyword, in every country, on both stores. One email at launch, with the early-bird price.