Connect an AI agent — Claude, Cursor, Claude Code, or anything else that speaks MCP — to the AppTweak API. The agent reads the full API reference, then either calls endpoints for you or writes code that does.
Explore the whole API without an API key
Browsing costs nothing and needs no credentials. The MCP exposes two kinds of tool, and only one of them touches AppTweak:
Documentation tools — list-specs, list-endpoints, search-endpoints, get-endpoint, search, fetch, start-here-guide — read our indexed OpenAPI specs. No API key, no credits, no AppTweak account. Your agent can learn every endpoint, parameter and response schema before you have a subscription.
The execution tool — execute-request — makes a real API call. This is the only tool that reaches AppTweak's backend, the only one that needs your API key, and the only one that spends credits.
Two practical consequences:
- You can evaluate the entire API surface with an agent, for free, and decide afterwards whether you need a plan.
- If you're building something, the agent can design the whole integration from documentation and never spend a credit until you run it. See Build with AppTweak data.
Tools
| Tool | What it does | Key needed | Credits |
|---|---|---|---|
start-here-guide | Spec map, known gotchas, and pointers to the two workflows that need special handling. Call this first. | No | 0 |
list-specs | Lists the three indexed OpenAPI specs | No | 0 |
list-endpoints | Every route in one spec | No | 0 |
search-endpoints | Keyword search across endpoints | No | 0 |
get-endpoint | Full detail for one endpoint: parameters, response schema, examples | No | 0 |
search / fetch | Search and retrieve any documentation page or reference | No | 0 |
app-id-lookup | Resolve a store URL or app name to an AppTweak app ID | No | 0 |
gpc-keyword-installs-guide | Walkthrough for Google Play Console installs-per-keyword with a branded / generic / competitor breakdown | No | 0 |
execute-request | Runs a real API request; returns status, body, and the credit cost | Yes | Yes |
The three specs
Point your agent at the right one and it will find endpoints far faster.
| Spec | Covers |
|---|---|
apptweak-store-api | The main ASO dataset — app metadata and metrics, category and keyword rankings, reviews, keyword metrics, top charts, paid keyword bids |
apptweak-integrations-api | Data from accounts you have connected — App Store Connect, Google Play Console. Requires the connection to exist first; a 200 with empty data usually means it doesn't. |
apptweak-api | Utility endpoints — your credit balance, valid countries and languages, DNA reference data, tracked-applications CRUD |
Connect
The server is at https://developers.apptweak.com/mcp. Authentication is one header, x-apptweak-key, holding your API key.
See Connect your client for per-client configuration.
MCP, direct API, or recipes?
Three ways to work with AppTweak data. They're complementary.
| Use it when | |
|---|---|
| MCP | You want an agent to explore the API, answer a question, or write integration code for you. Best for discovery and for one-off analysis. |
| Direct API | You know the endpoint and you're running it repeatedly: a dashboard refresh, a scheduled export, anything in CI. Cheaper and faster than routing through an agent. |
| Recipes | You want a working script for a known use case. Ready-made Python and Ruby samples you can paste into a notebook. |
The common pattern for builders is MCP then direct API: let the agent figure out the calls, then run those calls from your own code.
Costs and limits
- Documentation tools are free.
execute-requestspends credits from your API plan, exactly like a direct API call. Every response carriesmetadata.request.cost.- The API allows 60 requests per 10 seconds.
- An agent can fire many calls in one session, so credits deplete faster than they do with hand-written code. Read Credits and limits for agents before pointing an agent at a large job.
Before you build: read the gotchas
Call start-here-guide first, every session. It carries the current list of traps that will otherwise show up as wrong numbers in whatever you build — unreliable schema defaults, one deprecated endpoint that still appears in search results, and two endpoints whose default date ranges return empty. The same list is on Agent context files in a form you can paste into your repo.