REST or MCP
Call it from your code, or let an AI agent call it for you.
Upscrape has two doors to the same endpoints: the REST API for your code, and an MCP server for AI agents. Both use the same account, the same credits and the same prices, and return the same data.
Use REST when your code makes the call
Your program knows which endpoint it needs and calls it, the way it calls any other API.
A backend, a data pipeline, a scheduled job or a script.
You decide the retries, the storage and the schedule.
You want a typed client: every platform publishes an OpenAPI file.
Start with the Quickstart.
Use MCP when an AI agent makes the call
You ask in plain words, such as "find coffee shops in San Francisco with a rating over 4.5", and the agent finds the endpoint, checks its price and runs it.
Claude, ChatGPT, Cursor and other AI apps.
Questions you can't list in advance, where the agent picks the platform.
No key to paste: the app signs in with your Upscrape account.
Start with the MCP overview, or pick your app on the MCP server page.
What stays the same
| REST | MCP | |
|---|---|---|
| Endpoints | Every endpoint | Every endpoint |
| Price | Shown before the call | Shown to the agent before it runs |
| Sign-in | API key | Your account (OAuth), or an API key |
| Slow calls | Wait up to 30 s, then check back | Wait up to about 55 s, then check back |
| Failed calls | Free | Free |
Both at once
Many teams use both: an agent to explore the data and find the right endpoint, then the same endpoint over REST in production code. If an AI coding agent is writing that code, give it the REST contract; Build with a coding agent shows how.
Something missing or unclear? Ask the team that builds the endpoints.