generated: '2026-08-14'
method: derived
source: >-
  Derived by binding the three live MCP tools returned by
  POST https://docs.herondata.io/mcp (tools/list, HTTP 200) against the 272 operations in
  openapi/heron-openapi.json (Heron Data API, fetched from https://app.herondata.io/swagger).
description: >-
  Heron's MCP surface and its REST surface do NOT overlap. The MCP server is a
  documentation-retrieval server (search, read, feedback) hosted on docs.herondata.io;
  the REST API is the product surface on app.herondata.io. Every MCP tool is therefore
  mcp_only, and every one of the 272 REST operations is rest_only. This is the honest
  result and it is the useful one: an agent cannot drive Heron through MCP today, and
  the crosswalk records exactly that gap rather than inventing bindings.
surfaces:
  openapi:
    path: openapi/heron-openapi.json
    title: Heron Data API
    version: '2021-07-19'
    server: https://app.herondata.io
    auth: apiKey (x-api-key header)
    operations: 272
    gated: false
  mcp:
    url: https://docs.herondata.io/mcp
    tools: 3
    auth: none
    gated: false
    scope: documentation only
  graphql:
    present: false
crosswalk: []
mcp_only:
  - tool: search_heron
    reason: >-
      Documentation search over docs.herondata.io. There is no REST operation in the Heron
      Data API that searches documentation — the nearest REST search operations
      (/api/merchants/search, /api/funders/search) search product data, not docs, so they
      are not a binding.
  - tool: query_docs_filesystem_heron
    reason: >-
      Read-only virtual filesystem over documentation pages and specs. No REST equivalent;
      the operation it most resembles is fetching the published spec at /swagger, which is
      not an API operation.
  - tool: submit_feedback
    reason: >-
      Submits documentation feedback to the Heron docs team. Distinct from the product
      feedback endpoints (/api/enrichment_feedback/*), which record merchant and category
      corrections on transactions, not documentation defects.
rest_only_summary:
  count: 272
  note: >-
    All 272 REST operations lack an MCP tool. The highest-value unmapped flows — the ones
    an agent would need first — are listed below by tag, using operation paths from the
    published spec.
rest_only_highlights:
  - tag: EndUsers
    operations: 27
    example_paths:
      - POST /api/end_users
      - GET /api/end_users/{end_user_id_or_heron_id}
      - PUT /api/end_users/{end_user_id_or_heron_id}
      - POST /api/end_users/{end_user_id_or_heron_id}/enrich
  - tag: EndUserFiles
    operations: 22
    example_paths:
      - POST /api/end_users/{end_user_id_or_heron_id}/files/v2
      - GET /api/end_users/{end_user_id_or_heron_id}/files
  - tag: Workflows
    operations: 26
    example_paths:
      - POST /api/end_users/{end_user_id_or_heron_id}/start_workflow
  - tag: EndUserCalculations
    operations: 18
    example_paths:
      - GET /api/end_users/{end_user_id_or_heron_id}/scorecard
      - GET /api/end_users/{end_user_id_or_heron_id}/bank_statement_summary
  - tag: Webhooks
    operations: 5
  - tag: Transactions
    operations: 4
  - tag: Integrations
    operations: 13
coverage:
  mcp_tools: 3
  mcp_tools_bound_to_rest: 0
  rest_operations: 272
  rest_operations_with_tool: 0
  overlap_percent: 0
  confidence: high
  basis: >-
    Both surfaces were read live and unauthenticated — the MCP tool list from tools/list
    and the REST operations from the provider's own published OpenAPI — so the zero
    overlap is measured, not inferred.