generated: '2026-08-29'
method: derived
source: >-
  mcp/scrapingant-mcp-tools-list.json (probed 2026-08-29),
  openapi/_original/scrapingant-openapi.json (harvested 2026-08-29),
  https://docs.scrapingant.com/llm-markdown, https://docs.scrapingant.com/json-response,
  https://docs.scrapingant.com/ai-data-extraction/ai-extractor
provider: ScrapingAnt
providerId: scrapingant
description: >-
  Binds each ScrapingAnt MCP tool to the REST operation(s) that back it. The two surfaces are
  deliberately asymmetric: the MCP server exposes three output FORMATS of one underlying fetch,
  while the REST API exposes one operation per format-carrying PATH plus a usage/metering
  operation the MCP server does not surface at all. The published OpenAPI is also narrower than
  the documented REST surface - /v2/markdown, /v2/extract and /v2/extended are documented and
  live but absent from api.scrapingant.com/openapi.json, which is why two crosswalk rows bind
  to documented-but-unspecified operations and carry confidence: medium.
surfaces:
  openapi:
    file: openapi/_original/scrapingant-openapi.json
    url: https://api.scrapingant.com/openapi.json
    gated: false
    operations_declared: 6
    note: >-
      Six operation objects across two paths, but the document reuses one operationId
      (scrapingant_general_request_v2_general_post) across all five methods of /v2/general.
  mcp:
    url: https://api.scrapingant.com/mcp/
    gated: false
    note: >-
      tools/list answers anonymously with full inputSchema. Tool INVOCATION requires x-api-key.
    tools_declared: 3
  graphql:
    present: false
crosswalk:
- tool: get_web_page_html
  category: fetch
  rest:
  - scrapingant_general_request_v2_general_get
  binding: direct
  confidence: high
  note: >-
    Both return the raw HTML body of the rendered page. The tool's url/browser/proxy_type/
    proxy_country parameters are exactly four of the ten query parameters on GET /v2/general
    and carry identical names, types and defaults.
  inherits_input_schema_from: GET /v2/general
- tool: get_web_page_markdown
  category: fetch-transform
  rest:
  - GET /v2/markdown
  binding: direct
  confidence: medium
  note: >-
    docs.scrapingant.com/llm-markdown documents https://api.scrapingant.com/v2/markdown taking
    the same request structure as /v2/general and returning {"url","markdown"}. The endpoint is
    live (422 "API token is missing" when called without a key, same as the specified paths)
    but is NOT present in the published OpenAPI, so no operationId exists to bind to.
    Confidence is medium because the mapping is read from the docs, not the contract.
- tool: get_web_page_text
  category: fetch-transform
  rest: []
  binding: none
  confidence: medium
  note: >-
    No REST path returns plain text as its primary representation. The closest REST equivalent
    is the `text` property of the JSON body returned by the documented /v2/extended endpoint
    (docs.scrapingant.com/json-response), so a REST caller reaches the same data as a field
    rather than as a response body. Treated as a projection, not a binding.
mcp_only:
- tool: get_web_page_text
  reason: >-
    Plain-text extraction is exposed as a first-class tool on MCP but only as the `text` field
    inside the /v2/extended JSON envelope on REST. An agent gets it in one call; a REST client
    must request the extended envelope and read a field.
rest_only:
- operation: scrapingant_usage_v2_usage_get
  path: GET /v2/usage
  reason: >-
    Credit-balance metering. No MCP tool reports remaining credits, so an agent driving the MCP
    server cannot see how much of the plan quota it has consumed or how close it is to a 403.
    This is the single most consequential gap in the crosswalk for autonomous use.
- operation: POST /v2/general
  reason: >-
    /v2/general accepts POST, PUT, PATCH and DELETE and proxies that method through to the
    TARGET site. The MCP tools are fetch-only - no tool lets an agent issue a non-GET request
    against a target URL.
- operation: PUT /v2/general
  reason: See POST /v2/general - method pass-through is REST-only.
- operation: PATCH /v2/general
  reason: See POST /v2/general - method pass-through is REST-only.
- operation: DELETE /v2/general
  reason: See POST /v2/general - method pass-through is REST-only.
- operation: GET /v2/extract
  reason: >-
    AI structured-data extraction (extract_properties free-text schema -> JSON). Documented at
    docs.scrapingant.com/ai-data-extraction/ai-extractor, live, absent from the published
    OpenAPI, and NOT exposed as an MCP tool. An agent that wants typed JSON rather than page
    content has to fall back to the REST API.
- operation: GET /v2/extended
  reason: >-
    Extended JSON envelope (html, text, cookies, status_code, headers, xhrs, iframes).
    Documented, live, absent from the published OpenAPI, no MCP tool.
parameter_divergence:
- parameter: timeout
  rest: documented (5-60s, default 60)
  mcp: absent
  note: >-
    An agent cannot bound how long a ScrapingAnt MCP call may run. Documented on the REST
    surface at docs.scrapingant.com/request-response-format; not in the tool inputSchema.
- parameter: return_page_source
  rest: present
  mcp: absent
  note: The cheap no-JS-rendering path (2 credits vs 10) is unreachable from MCP.
- parameter: js_snippet
  rest: present
  mcp: absent
- parameter: cookies
  rest: present
  mcp: absent
- parameter: wait_for_selector
  rest: present
  mcp: absent
- parameter: block_resource
  rest: present
  mcp: absent
coverage:
  mcp_tools: 3
  mcp_tools_bound_to_rest: 2
  mcp_only: 1
  rest_operations_specified: 6
  rest_operations_documented_but_unspecified: 3
  rest_only: 8
  bound_high_confidence: 1
  bound_medium_confidence: 1
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com