An agent using the MCP spends credits from your API plan exactly as your own code would. The difference is that an agent decides how many calls to make, and it will happily make more than you expected. This page is how to stay in control.
What costs, and what doesn't
| Credits | |
|---|---|
Documentation tools (search-endpoints, get-endpoint, list-specs, start-here-guide, app-id-lookup, the guide tools) | None. They read our indexed docs and never reach the API. No key required either. |
execute-request | Yes. Identical to calling the endpoint directly. |
So exploration is free and unlimited. Only execution bills.
How a request is priced
Three principles drive the model:
- Current data costs 1 credit per app and per metric; history costs 5, for any date range up to 1 year.
- Download and revenue estimates cost 10 credits per app (current) or 50 (history), on Enterprise plans with estimates access and on existing plans that retain access. New Lite, Pro and Power plans do not include estimates.
- The price is known from the parameters: it does not depend on what the API returns.
What that means in practice
A few reference points, so nothing surprises you:
| Request | Credits |
|---|---|
| App metadata, 1 app | 1 credit |
| Downloads, 1 app, current | 10 credits |
| Downloads, 1 app, 45 days | 50 credits |
| Category downloads, 1 category, 30 days | 50 credits |
Why agents spend faster than people
A developer writing code makes one considered call. An agent in a conversation may:
- retry a call after misreading a parameter,
- fetch the same range twice while refining an analysis,
- fan out across apps or keywords one call at a time instead of batching,
- loop day-by-day instead of asking for a range.
None of this is misbehavior; it's how an exploratory loop works. It's also why the cost of an agent session is best measured rather than predicted.
Staying in control
Check the cost on every response.
"metadata": {
"request": {
"cost": 5,
"charge_type": "credits",
"status": 200
}
}Your remaining balance isn't in the response — see below.
Request a whole date range (up to 1 year) in one call: it costs the same as a single day.
Batch what batches. Most endpoints accept several apps or keywords per call. Credits are charged according to the endpoint’s price and the apps and metrics requested. Keyword rankings accept up to five keywords per call.
Tell your agent the budget. Instructions like "check the documented credit cost before executing, and ask me before any single call over 100 credits" work well and cost nothing to add. Put it in your AGENTS.md and it applies every session.
Have the agent check the balance — it's free. Remaining credits aren't part of the response envelope, so it takes a separate call: the usage/credits endpoint on apptweak-api. That endpoint costs nothing, so there's no reason not to call it at the start of a heavy session, or between batches. Ask your agent to check the balance first and report what it has to work with.
Move repeat work off the agent. Anything you run more than once belongs in a script calling the API directly — same credit cost, no agent overhead, no risk of an unplanned retry. See Build with AppTweak data.
Rate limit
60 requests per 10 seconds. An agent fanning out across many apps can hit this; so can a generated script with an uncapped loop. Build in a small delay or a retry with backoff.
Plans
MCP access needs an API plan — the MCP calls the same metered API. Plans and their credit allowances are on the pricing page; a 7-day free trial is available. Your key and current balance live in your API dashboard.
Remember that everything except execute-request is free and keyless, so you can have an agent explore the whole API before you subscribe.