How to Use the GetAppNiche MCP with Claude for App Research

This is a hands-on tutorial: by the end, Claude will answer app-market questions — “which of these three competitors is actually growing?”, “is this niche worth entering?” — by calling live GetAppNiche data itself, across 3.6M+ apps on the App Store and Google Play, with revenue and download figures clearly labeled as estimates. If you first want the conceptual case for wiring an AI assistant to app data at all, read the companion explainer on App Store MCP servers; this page assumes you’re sold and just want it running.

The short video below shows it working in Claude Code: one command to connect, then real calls that find a niche, score keywords and check what a rival changed.

2:03

Using Claude in the browser? You don’t need an API key — Step 1 connects claude.ai by signing in to your GetAppNiche account. Claude Code and the config-file route in Claude Desktop need a GetAppNiche API key instead. Create it in the app under Settings → API Keys (sign in and continue to API Keys) — the secret is shown once and looks like getappniche_....

Step 1 — Connect claude.ai (sign in, no API key)

The fastest way: GetAppNiche is listed in Claude’s connector directory. Open GetAppNiche in the directory, click Connect, sign in to GetAppNiche and choose Allow access. No key to copy, no config file.

You can also add the same server by hand as a custom connector — paste the URL once, sign in, and Claude keeps the connection. Custom connectors work on every Claude plan; the free plan allows one.

1. In Claude, open Connectors (under Customize or Settings — Claude moves this menu now and then) and choose Add custom connector.
2. Name it GetAppNiche and paste https://api.getappniche.com/mcp as the MCP server URL.
3. Under Authentication, choose Sign in now. Claude may preselect No sign-in for this server — switch it, or the tools will ask you to sign in and stop. Keep the recommended OAuth client option.
4. Click Add, then Connect. Sign in to GetAppNiche and choose Allow access.
5. In a new chat, ask your question. Claude asks before each tool call until you allow the tools in the connector’s settings.

The connector can read market data and nothing else: it can’t change your account, billing, API keys or tracked apps. Disconnect it any time in GetAppNiche under Settings → Connected apps. The same steps work in the Claude desktop app if you’d rather sign in than edit the config file in Step 3.

Step 2 — Connect Claude Code

The server is hosted at https://api.getappniche.com/mcp (JSON-RPC 2.0 over Streamable HTTP), authenticated with Authorization: Bearer <your key>. In Claude Code, register it as a remote HTTP server:

claude mcp add --transport http getappniche https://api.getappniche.com/mcp --header "Authorization: Bearer YOUR_API_KEY"

To verify, start a session and ask: “What GetAppNiche tools do you have?” Claude should list eight: search_apps, get_app_detail, get_app_historicals, get_app_timeline, get_keyword_difficulty, batch_keyword_difficulty, get_app_reviews, get_supported_countries. If it lists nothing, the two usual culprits are a mistyped key (a 401 means the header didn’t carry the full getappniche_... secret) or a wrong URL — the endpoint is exactly /mcp, no trailing path.

Step 3 — Connect Claude Desktop

Claude Desktop reaches the same server through a small config entry (found in Claude’s connector/developer settings — the config file is claude_desktop_config.json). It uses the official @getappniche/mcp bridge — zero dependencies, your key travels only as a TLS Authorization header (requires Node 18+):

{
  "mcpServers": {
    "getappniche": {
      "command": "npx",
      "args": ["-y", "@getappniche/mcp"],
      "env": { "GETAPPNICHE_API_KEY": "YOUR_API_KEY" }
    }
  }
}

Restart Claude Desktop after saving; the GetAppNiche tools appear in the tools menu. Exact per-client snippets (including Cursor and other Streamable-HTTP clients) live in the MCP setup page.

Step 4 — Run real research workflows

You never call tools by name — you ask questions and Claude picks. But knowing which tools a prompt will trigger tells you what it costs. Three workflows worth stealing:

Workflow 1: Competitor brief

Use GetAppNiche tools to pull revenue estimates and the 90-day trend for Headspace, Calm, and Balance. Summarize in one page: how each monetizes, which is growing, and where the gap is. Treat all revenue figures as estimates.

Claude chains get_app_detail (1 credit each) and get_app_historicals (1 credit each) per app — about 6 credits total, plus a search_apps call or two (1 credit per app returned, 50 a call unless Claude asks for fewer) if it needs to resolve names to app IDs. You get a brief grounded in numbers Claude just fetched, not remembered. Add “and what each changed in its store listing recently” and Claude calls get_app_timeline per app (1 credit each): new versions, price and business-model changes, renames, category moves and icon or screenshot redesigns, grouped by day.

Workflow 2: Niche scan

Use search_apps to find iOS apps matching “habit tracker” with estimated revenue over $3,000/month. Show me the estimated revenue distribution — how many small, mid, and large earners — and give a verdict: is there room for a new entrant?

One or a few search_apps calls (1 credit per app returned — 50 a call by default, anywhere from 1 to 100 if you name a page size) with follow-up filters as Claude narrows. Add “then score ‘habit tracker’ and 4 related keywords for difficulty” and it calls batch_keyword_difficulty — that’s the expensive one at 10 credits per keyword, so a 5-keyword add-on costs 50. Budget accordingly: three searches at the default 50 rows plus five keywords is about 200 credits — ask for 10-row pages and it drops to about 80.

Workflow 3: Review mining

Pull recent reviews for the top habit tracker from the last search. Cluster the recurring complaints into themes, and list 3 feature gaps a new app could win on.

get_app_reviews costs 1 credit per review returned (50 a call by default with an API key, 20 when an assistant connects by signing in; up to 100), so mining three competitors at the default 50 reviews each is about 150 credits — or 60 if you ask for 20 each. This is the conversational version of the manual process in our app review analysis guide — same signal, no spreadsheet stage.

Credits and limits

• 5,000 API credits per month on Pro, auto-refreshed; every tool response reports credits_charged. Extra 500-credit packs are available in the app.
• 60 requests per minute per key or connection — a 429 tells Claude to slow down and retry.
• Read-only. The tools query market data; they can’t modify anything in your account or touch your files.

At 1 credit per row for searches and reviews and 1 per call for detail, 5,000 credits is a few dozen full research sessions a month. Keyword difficulty (10 per keyword) and big pages are what burn fast — tell Claude the page size you actually need (“top 10”, “20 reviews”).

What this setup won’t do

Honest boundaries, so you don’t build a workflow on sand:

• Estimates, not actuals. Revenue and download figures are modeled from public store signals. Tell Claude to compare and rank with them, not to quote them as financials.
• Store coverage. App search, app detail, the listing timeline, keyword scoring and review tools all cover both the App Store and Google Play — pass a package name (google:com.duolingo) exactly where you would pass an iOS id. Chart rankings cover both stores in the app, but no MCP tool serves them.
• No rank tracking through MCP. There is no rank-history tool (get_app_timeline records listing changes, not search positions), so Claude cannot pull chart positions or keyword-rank timelines from this server — don’t prompt for them, you’ll get hallucination risk instead of data.
• Daily refresh. Data updates daily, so “live” means today’s snapshot, not minute-level.

Start with your real question

The best first prompt is the research question already on your list. Connect the server, paste one of the workflows above, and let Claude do the pulling — subscribe to Pro for the 5,000-credit pool that powers it, then sign in from claude.ai or create an API key for Claude Code.