ASO with Claude and ChatGPT: workflows with an MCP server
To do ASO with Claude or ChatGPT, connect the assistant to your real store data through an MCP server. With the Rankglass MCP server and a workspace API key, the assistant reads your tracked ranks, rank changes, competitor keywords, ASO score, weekly brief and keyword game plans, and answers from that data instead of from memory.
Without that connection, an assistant can explain ASO in general, but it can't know where your app ranks today, what your competitors changed last week or which of your keywords dropped. It will either say so or, worse, guess. This guide shows how to connect it, which prompts work well, and where the limits are.
What is an MCP server, and why use one for ASO?
The Model Context Protocol (MCP) is an open standard that lets AI assistants call tools on other systems. An MCP server describes its tools, the assistant picks the ones a question needs, and the server returns data the assistant then reasons over.
For ASO that matters for three reasons:
- Store data changes daily. Ranks move, competitors edit their titles, reviews come in. A model's training data can't know any of it.
- The useful questions combine sources. "Why did this keyword drop?" needs your ranks, your competitors' store page changes and the dates, side by side.
- You already work there. If you write code with Claude Code or Codex, asking about your ranks in the same window saves a trip to a dashboard.
The Rankglass MCP server is remote (it runs on our servers, not on your computer), uses the Streamable HTTP transport, and only reads. It is part of the Starter, Pro and Agency plans.
Which assistants can connect?
| Assistant | How it connects |
|---|---|
| Claude Code | One terminal command (claude mcp add) |
| Claude Desktop | claude_desktop_config.json, through mcp-remote (needs Node.js) |
| Cursor | ~/.cursor/mcp.json |
| Codex CLI, Codex IDE extension and the ChatGPT desktop app | ~/.codex/config.toml, which OpenAI says the three share |
| Any other MCP client | The server URL and an Authorization header |
ChatGPT on the web works differently: according to OpenAI's MCP documentation, it uses remote MCP tools through plugins, a setup this guide doesn't cover. The desktop app and Codex can send the API key directly, as shown below.
How do you connect Rankglass?
First create a key: in Rankglass, open Settings → Workspace → API & MCP and click Create key. Workspace owners and admins can do this. The key is shown once, so copy it right away. Then add the server, replacing YOUR_API_KEY with the key.
Claude Code (Anthropic's MCP docs):
claude mcp add --transport http --scope user rankglass https://app.rankglass.com/api/mcp \
--header "Authorization: Bearer YOUR_API_KEY"
Codex and the ChatGPT desktop app, in ~/.codex/config.toml, with the key (without Bearer) in an environment variable named RANKGLASS_API_KEY:
[mcp_servers.rankglass]
url = "https://app.rankglass.com/api/mcp"
bearer_token_env_var = "RANKGLASS_API_KEY"
OpenAI's documentation also lists http_headers, which sets a fixed header if you would rather keep the key in the file.
The snippets for Cursor and Claude Desktop are in our setup post, and the MCP page lists every tool. Any other client needs three things: the URL https://app.rankglass.com/api/mcp, the Streamable HTTP transport and the header Authorization: Bearer YOUR_API_KEY (or x-api-key: YOUR_API_KEY if it can't set Authorization).
To check that it works, ask: "Which apps are in my Rankglass workspace?" The assistant should call list_apps and name your apps, their stores and tracked countries.
What can the assistant read?
Eleven tools, each with a matching REST endpoint:
list_apps: your apps, stores, tracked countries and this week's headline numbersget_keyword_rankings: every tracked keyword in one store and country, with rank, 7-day change, best rank, estimated monthly searches, difficulty and competitor ranksget_rank_changes: the biggest gains and losses over 1 to 29 days, and keywords that entered or left the top 10get_competitors: followed competitors with ratings, head-to-head counts and keyword gapsget_competitor_keywords: what a competitor ranks for in one country, including keywords you don't track yetget_competitor_insights: a competitor's store facts, chart ranks, estimated downloads and revenue, and its latest review summaryget_competitor_growth: how a competitor gets its users, with estimates and a short AI strategy (Pro and Agency)get_store_page_changes: recent changes to your store page and your competitors' pagesget_aso_score: your ASO score with its checks, quick keyword wins and the latest AI suggestionsget_weekly_brief: the latest weekly brief: what changed, likely reasons and next stepsget_keyword_game_plan: the latest game plan for a tracked keyword: the goal rank and the steps to reach it
You don't need to name the tools in your prompts; the assistant picks them from your question. Naming them helps when you want a specific one.
Which prompts work well for ASO?
The prompts below use a sample app called Habitly. Replace the app, keywords, store and country with yours. Each one names the store and country, because ranks are different in every storefront.
1. The Monday check-in
Using Rankglass, get the latest weekly brief for Habitly and the rank changes of the last 7 days. Give me five bullet points: what moved, the likely reason and the one thing to do this week. Name the store and country for every keyword.
Tools: get_weekly_brief, get_rank_changes. On paid plans, the weekly brief is written every Monday for each app with at least three tracked keywords, so this is a good first question of the week. If the brief and the rank changes disagree, ask the assistant to show both.
2. Why did a keyword drop?
"habit tracker" dropped for Habitly on the App Store in the US this week. Using Rankglass, check my rank changes, my keyword rankings for that store and country, and the store page changes of the last 14 days for my app and my competitors. Tell me what changed around the day it dropped. Separate what the data shows from what you are guessing.
Tools: get_rank_changes, get_keyword_rankings, get_store_page_changes. The last sentence matters: a competitor adding "habit tracker" to its title the day before your drop is evidence, not proof, and a good answer says so.
3. Find keywords a competitor ranks for
On the App Store in Germany, list Habitly's competitors, then get the keywords the strongest one ranks for that I don't track yet. Show the 15 with the most monthly searches where it is in the top 10, with difficulty, and mark the ones that fit my app.
Tools: get_competitors, get_competitor_keywords (its filter is untracked by default). In the US it returns the competitor's full keyword list, which Rankglass reads (at most once a week) when you open the competitor's Keywords tab in the dashboard, so open that tab once first: until then, the US list is empty or holds only the few keywords Rankglass's own searches found. In the other countries, the list holds what Rankglass's keyword searches found, so run Find keywords on that tab first. How to find a competitor's App Store keywords in any country explains the method behind it.
4. Rewrite the subtitle and keyword field
Get Habitly's ASO score for the App Store in the US and its keyword rankings there. Suggest three subtitles of at most 30 characters and one keyword field of at most 100 bytes: terms separated by commas with no spaces, no words already in my title or subtitle, no plurals of words already there, no competitor names. Use only keywords from my tracked list or the quick wins, and say which keyword each suggestion is for.
Tools: get_aso_score, get_keyword_rankings. The rules in the prompt come from Apple's product page guide. Language models are unreliable at counting characters and bytes, so paste the result into the free listing checker before you use it: it counts each field against Apple's limits and flags plurals, duplicates and the word "app".
5. Turn a game plan into a release checklist
Get the game plan for "daily planner" for Habitly on the App Store in the US. Turn its steps into a checklist for my next release, and mark which steps need a new app version and which I can do today.
Tool: get_keyword_game_plan. Apple says you change the subtitle when you submit a new version, while promotional text can be updated at any time, so the split is real. Game plans are written in the dashboard, at most once a week per keyword; if there is none yet, open the keyword's game plan in Rankglass first.
6. What do a competitor's users complain about?
Get the competitor insights for Competitor A on the App Store in the UK. Summarize what its reviewers complain about and ask for, and list three things in Habitly that answer those complaints. Keep downloads and revenue as the ranges Rankglass gives and call them estimates.
Tool: get_competitor_insights. Review summaries are made in the dashboard (Analyze reviews on the competitor's Reviews tab) and quote real reviews. The stores don't publish downloads or revenue, so those figures are always estimates; asking the assistant to keep them as ranges stops it from turning "3K–6K" into "about 4,500".
7. How does a competitor get its users?
Get the growth research for Competitor A. Which channels is it active on, which aren't visible from outside, and what are the moves in its strategy summary? Keep every spend and install figure as the range Rankglass gives, labeled as an estimate.
Tool: get_competitor_growth, on the Pro and Agency plans. It returns the last growth research you ran from the competitor's Growth tab (Beta), and nothing until you run one.
8. A weekly report in your own tools
In Claude Code or Codex, the assistant can also write code against the REST API:
Write a script that reads the rank changes of the last 7 days from https://app.rankglass.com/api/v1/rank-changes?days=7, with the API key from the RANKGLASS_API_KEY environment variable in an Authorization: Bearer header, and writes the five biggest gains and drops per app to a Markdown file.
The REST API returns the same data as the MCP tools, as JSON. Its OpenAPI document describes every endpoint, so you can point the assistant at it.
What are the limits?
- It only reads. The assistant can't add keywords, follow competitors or change anything in your workspace. It also can't start paid work: keyword searches for a competitor, review summaries, game plans and growth research are started in the dashboard, and the tools read what they found.
- Data is as fresh as the last check. On paid plans, each tracked keyword is checked once a day, in each store and country.
- Rate limits. Each API key can make 60 requests a minute and 2,000 a day, counted together across the REST API (
/api/v1) and the MCP server (/api/mcp). Every request to either counts: each tool call, and also the requests an MCP client sends on its own, such as connecting and listing the tools. Past the limit, requests get a429response with aRetry-Afterheader. - Plans. API and MCP access come with the Starter, Pro and Agency plans, not the Free plan. The Growth tools need Pro or Agency. Current prices and limits are on the pricing page.
- Keys. A workspace can have up to 20 keys, and a key can read everything in its workspace. Make one per assistant or script so you can revoke each on its own.
- Countries. Rankglass tracks the App Store and Google Play in 40 countries. Questions about other countries get no data.
- Estimates stay estimates. Search volumes, downloads, revenue, ad spend and installs from ads are estimates, because the stores don't publish them. Ask the assistant to label them that way.
- Long answers cost tokens. A full competitor keyword list can hold hundreds of rows. Ask for a number ("the top 15") and the assistant can pass a
limit.
How do you get better answers?
- Name the app, store and country in every question. Without them, the tools use your app's main country, which may not be the one you meant.
- Ask for numbers with their source. "Show the ranks you used" makes it easy to spot a wrong reading.
- Separate data from guesses. Ask the assistant to mark which statements come from Rankglass and which are its own reasoning.
- Check text before you ship it. Counts, plurals and duplicates are easy for the listing checker and easy for a model to get wrong.
- Keep the decision yours. The assistant reads the same data the dashboard shows. It is fast at combining it, but it doesn't know your roadmap, your brand or what you tried last year.
Sources
Checked on 16 November 2026.
- Model Context Protocol
- Anthropic: Connect Claude Code to tools via MCP
- OpenAI: Model Context Protocol
- Apple Developer: Creating your product page
- Rankglass: MCP server and REST API and setup post