Start

Documentation

Connect AI clients to live platform data through MCP.

Upscrape gives AI clients and applications one live catalog of structured platform capabilities. Connect the MCP server when an AI client should discover and call tools. Use REST when your application owns the execution flow.

> **For AI agents.** Start with [`llms.txt`](/llms.txt) for discovery, fetch [`llms-full.txt`](/llms-full.txt) for the complete implemented documentation, or request any documentation URL with `Accept: text/markdown`.

> **Start with the contract.** Every public platform page publishes its capability IDs, input schemas, example requests, credit cost, and, when safe, a real redacted sample response.

## Two ways to connect

Use **MCP** for AI clients and agents. The client searches the live catalog, reads the selected capability's exact schema, executes it, and retrieves a pending result when needed. Start with the [MCP overview](/docs/mcp/overview).

Use **REST** when your application owns control flow, retries, persistence, and scheduling. Start with the [five-minute quickstart](/docs/quickstart).

## The execution model

Every REST capability uses `POST /execute`. A request identifies a capability and supplies input matching that capability's JSON Schema. Upscrape validates the request, resolves account and network policy, runs the registered worker, and returns structured JSON.

Fast work can complete in the initial response when you request a bounded wait. Otherwise, Upscrape returns a job ID that you poll until it completes or fails.

MCP uses the same catalog and execution system as REST. Authentication, visibility, credit accounting, idempotency, and capability behavior stay consistent across both integrations.

## Authoritative references

- [Browse platforms and capabilities](/scrapers)
- [Copy the complete capability catalog as Markdown](/scrapers/llm.md)
- [Understand REST authentication](/docs/api/authentication)
- [Connect an MCP client with OAuth](/docs/mcp/oauth)
- [Review result provenance and trust boundaries](/docs/reference/provenance)

## What is not available yet

Planned documentation pages are visible in the navigation and marked **Planned**. They describe the intended boundary without pretending that the feature exists. The implementation work is tracked in the repository's `asap-todo.md`.

Set up Upscrape via MCP

One hosted server. OAuth by default. Works across leading AI clients.

See all MCP guides

Add Upscrape as a custom connector

Open Claude Settings, choose Connectors, then add this remote server URL.

https://data.upscrape.com/mcp

Authorize the connection

Claude opens Upscrape in your browser. Sign in, review access, and approve the connector.

Open the OAuth setup guide

Run this command in your terminal

This registers Upscrape as a remote MCP server in Codex.

codex mcp add upscrape --url https://data.upscrape.com/mcp

Sign in and verify

Run codex mcp login upscrape, complete OAuth, then use codex mcp list to confirm the server.

Review Codex authentication

Add the Upscrape connector

Open ChatGPT connector settings and create a custom connector with this server URL.

https://data.upscrape.com/mcp

Sign in with Upscrape

ChatGPT follows the server metadata and opens the Upscrape OAuth consent flow. No API key is copied into ChatGPT.

Open the connector setup guide

Four tools, the full catalog. Search capabilities, inspect schemas, execute work, and retrieve long-running results without loading hundreds of schemas into context. See the tool model

Upscrape gives AI clients and applications one live catalog of structured platform capabilities. Connect the MCP server when an AI client should discover and call tools. Use REST when your application owns the execution flow.

For AI agents. Start with llms.txt for discovery, fetch llms-full.txt for the complete implemented documentation, or request any documentation URL with Accept: text/markdown.

Start with the contract. Every public platform page publishes its capability IDs, input schemas, example requests, credit cost, and, when safe, a real redacted sample response.

Two ways to connect

Use MCP for AI clients and agents. The client searches the live catalog, reads the selected capability's exact schema, executes it, and retrieves a pending result when needed. Start with the MCP overview.

Use REST when your application owns control flow, retries, persistence, and scheduling. Start with the five-minute quickstart.

The execution model

Every REST capability uses POST /execute. A request identifies a capability and supplies input matching that capability's JSON Schema. Upscrape validates the request, resolves account and network policy, runs the registered worker, and returns structured JSON.

Fast work can complete in the initial response when you request a bounded wait. Otherwise, Upscrape returns a job ID that you poll until it completes or fails.

MCP uses the same catalog and execution system as REST. Authentication, visibility, credit accounting, idempotency, and capability behavior stay consistent across both integrations.

Authoritative references

What is not available yet

Planned documentation pages are visible in the navigation and marked Planned. They describe the intended boundary without pretending that the feature exists. The implementation work is tracked in the repository's asap-todo.md.