generated: '2026-08-09'
method: derived
source: mcp/webscraping-ai-mcp.yml + openapi/webscraping-ai-openapi.yml
description: >-
  Binds each WebScraping.AI MCP tool to the OpenAPI operation that backs it. WebScraping.AI is
  an unusually clean case: the MCP surface is a 1:1 projection of the REST surface — seven tools,
  seven operations, no MCP-only composites and no REST operations without a tool. Each tool's
  real input contract is the bound operation's `parameters` block in
  openapi/webscraping-ai-openapi.yml.
surfaces:
  openapi:
    file: openapi/webscraping-ai-openapi.yml
    source: https://webscraping.ai/openapi.yml
    version: 3.2.1
    server: https://api.webscraping.ai
    gated: false
  mcp:
    url: https://mcp.webscraping.ai/mcp
    transport: streamable-http
    gated: true
    gated_note: >-
      tools/list requires an OAuth 2.1 bearer token (401 anonymously). Tool names and input
      shapes were taken from the provider's open-source stdio server, which the docs state
      exposes the same 7 tools.
  graphql: null
  proxy_mode:
    endpoint: proxy.webscraping.ai:8888
    note: >-
      HTTP proxy front-end for the same scraping core; not modeled in the OpenAPI and not
      exposed as an MCP tool.
crosswalk:
- tool: webscraping_ai_question
  category: ai
  rest: [getQuestion]
  path: GET /ai/question
  binding: rest
  confidence: high
  note: Tool args url+question map to the `url` and `question` query parameters.
- tool: webscraping_ai_fields
  category: ai
  rest: [getFields]
  path: GET /ai/fields
  binding: rest
  confidence: high
  note: >-
    Tool arg `fields` (a name -> instruction map) maps to the deepObject `fields` query
    parameter. The tool JSON-stringifies the result; REST returns the object directly.
- tool: webscraping_ai_html
  category: scraping
  rest: [getHTML]
  path: GET /html
  binding: rest
  confidence: high
  note: >-
    The tool adds a client-side `format` (json|text) wrapper that has no REST equivalent;
    REST always returns raw HTML.
- tool: webscraping_ai_text
  category: scraping
  rest: [getText]
  path: GET /text
  binding: rest
  confidence: high
  note: Tool args text_format and return_links map to the same-named query parameters.
- tool: webscraping_ai_selected
  category: scraping
  rest: [getSelected]
  path: GET /selected
  binding: rest
  confidence: high
  note: Tool arg `selector` maps to the `selector` query parameter.
- tool: webscraping_ai_selected_multiple
  category: scraping
  rest: [getSelectedMultiple]
  path: GET /selected-multiple
  binding: rest
  confidence: high
  note: Tool arg `selectors` (array) maps to the `selectors` array query parameter.
- tool: webscraping_ai_account
  category: account
  rest: [account]
  path: GET /account
  binding: rest
  confidence: high
mcp_only: []
rest_only: []
divergences:
- kind: response-shaping
  detail: >-
    The MCP server sanitizes and truncates page content before returning it, and wraps HTML /
    selected results in a JSON envelope when format=json. The REST operations return the raw
    payload. Same data core, different presentation.
- kind: auth
  detail: >-
    REST authenticates with an `api_key` query parameter; the hosted MCP server authenticates
    with OAuth 2.1 and issues no API key. The local stdio MCP server uses the API key.
- kind: parameter-defaults
  detail: >-
    The MCP server applies its own defaults (js rendering and residential proxy configurable via
    environment variables) on top of the REST defaults (js=true, proxy=datacenter).
coverage:
  tools_named: 7
  tools_bound: 7
  mcp_only: 0
  rest_operations_total: 7
  rest_operations_with_tool: 7
  binding_rate: 1.0