ASO from Claude Code: keyword research through an MCP server

Claude Code can do App Store keyword research once you give it a source of store data. A language model on its own has no live ranks and no demand numbers, so it guesses. VisualASO is a Mac app and an MCP server with 69 tools: connect it with one line in Terminal, and your agent reads ranks, demand and difficulty, and drafts listings.

Why an agent needs store data

An agent that writes your keyword field from its training data is writing from memory. It has no rank for your app in Germany this week, no demand number for a phrase, no idea which competitors hold the top ten. Connect it to a data source and the same agent works from what the store returns today, in each country you sell in.

The split matters because the two halves of ASO need different things. Reading the store — searching a phrase in a storefront, recording where you land, counting how strong the apps above you are — is mechanical work a program does well. Choosing which 100 characters go in the keyword field, and writing a subtitle that reads like English, is work a model does well. An MCP server hands the first half to the app and leaves the second half to the agent.

What MCP is, in one paragraph

MCP is the Model Context Protocol: a standard way for an agent such as Claude Code or Codex CLI to call tools that an app exposes. The app says what its tools are and what arguments they take; the agent picks one and reads the result. You register the server once and every later session can use it. The protocol has its own site if you want the specification.

Connect VisualASO to Claude Code

One line in Terminal, run once:

claude mcp add --scope user visualaso -- "/Applications/VisualASO.app/Contents/MacOS/VisualASO" --mcp

In the app, Help → Copy MCP Install Command copies that line with the path of the copy you are running, which saves trouble if VisualASO lives somewhere other than /Applications. Start a new Claude Code session afterwards and ask it to run status; that tool lists your tracked apps, markets, connections and recent jobs, and it is the right place to start a session. If you do not have the app yet, it is on the download page.

Agent tools are part of Pro. The free tier covers the app itself for 1 app in 3 countries with 25 keywords; Pro is one payment, and the pricing page has the detail. The full list of all 69 tools, with arguments, is in the agents and MCP docs.

Connect it to Codex

Codex CLI takes the same server. Either add it from the command line:

codex mcp add visualaso -- "/Applications/VisualASO.app/Contents/MacOS/VisualASO" --mcp

or write it into ~/.codex/config.toml:

[mcp_servers.visualaso]
command = "/Applications/VisualASO.app/Contents/MacOS/VisualASO"
args = ["--mcp"]

Nothing else differs. The tools, the data and the publishing lock are the same whichever agent calls them, because both talk to the same VisualASO data on your Mac.

There is a second, separate connection worth knowing about. VisualASO uses an AI tool you already have — Claude Code, Codex CLI or a local Ollama model — for its own writing, so the app drafts listing text with the plan you already have. You sign in to those tools yourself and VisualASO adds no AI fees. That is set up in the AI setup docs, and it is independent of whether you drive the app from an agent.

What to ask: a first session

Paste these in order. Each one is a plain sentence; the agent picks the tools. Name the server in the first prompt ("use visualaso") so the agent reaches for it instead of answering from memory.

1. "Use visualaso: show status, then add my app <App Store URL> for the US, UK and Germany." — status reports what is already tracked, app_add starts tracking the app from its URL or ID.
2. "Run a check and wait for it to finish, then show the board." — check_run searches every keyword in each market for rank, demand and difficulty; board_get returns the grid with a verdict per market.
3. "List the opportunities with demand above 30 where I am not in the top 10." — opportunities ranks starting points by high demand, low difficulty and few strong incumbents; the agent filters the list for you.
4. "Brainstorm 20 more phrases for this app and check them." — keywords_brainstorm rebuilds the list from your listing plus fresh ideas, then checks the new phrases.
5. "Propose three name, subtitle and keyword field options for the US, and explain each." — audit_english returns three main-listing options in your main locale, and the agent argues for each.
6. "Plan a probe for my localizations." Then, after your next version is live: "Run the probe and show which storefronts read each localization." — probe_plan plans one unique word per localization plus negative controls, and probe_run reads which storefronts rank each planted word.

Here are some of the 69 tools, so you can ask for one by name when you know what you want:

Tool	What it does
status	Tracked apps, markets, connections and recent jobs. Start here.
app_add	Starts tracking an app from its App Store URL or ID.
check_run	Searches every keyword in each market for rank, demand and difficulty.
board_get	The keyword grid with ranks, demand, difficulty and verdicts per market.
opportunities	Ranks the best starting points: high demand, low difficulty, few strong incumbents.
keywords_brainstorm	Rebuilds the keyword list from the listing plus ideas, then checks new phrases.
probe_plan / probe_run	Plans a probe (one unique word per localization, plus negative controls) and reads which storefronts rank each planted word.
audit_english	Proposes three main-listing options in the main locale.
appleads_popularity	Apple's search popularity for terms in one storefront, read from your own Apple Ads account.
changes_get	The listing change log with 21-day verdicts.

Two of those repay a closer look. Apple's own search popularity numbers come from your Apple Ads account, and our guide to Apple's popularity score explains how to read them. Probes exist because a storefront does not always read the localization you expect; our cross-localization research covers what that means for where you put your keywords.

What an agent can and cannot do

An agent can read and change your local VisualASO data, start checks and draft listings — everything the app itself does. Publishing is off. It can write to App Store Connect only after you turn on "Allow agents to publish" for that app in Optimize, and even then it starts as a dry run that writes nothing. No agent can submit a version for review; VisualASO cannot do that either.

The app and the server run at the same time, so you can watch the board fill while the agent works, and anything it changes shows up in the app. Keep "Allow agents to publish" off until you trust the drafts you are getting. Read every listing yourself before you create a version. You know your product better than any agent. The work happens on your own Mac, on your local data and your own Apple Ads keys.

Several ASO products now offer MCP servers, and they differ in ways worth checking before you wire one in: what store data it actually reads, which countries it reads it in, and whether it can write to App Store Connect without asking you first. That last one is the question to ask first.

Questions

Do I need Pro for MCP?

Yes — agent tools are part of Pro, which is one payment. The free tier covers the app itself for 1 app in 3 countries with 25 keywords.

Can the agent publish my listing?

Not until you allow it. Publishing is off by default, you turn on "Allow agents to publish" per app in Optimize, and the first write after that is a dry run that changes nothing. Submitting a version for review stays with you; neither the agent nor VisualASO can do it.

Does it work with Codex?

Yes. codex mcp add visualaso -- "/Applications/VisualASO.app/Contents/MacOS/VisualASO" --mcp, or the same command and --mcp argument written into ~/.codex/config.toml. The tools behave identically.

Do I pay extra for the AI that writes the drafts?

No. VisualASO calls an AI tool you already have — Claude Code, Codex CLI or a local Ollama model — and you sign in to that tool yourself, so VisualASO adds no AI fees.

Can I keep using the app while an agent works?

Yes. The app and the MCP server run at the same time, and changes an agent makes appear in the app.