specification: API Commons Tool Crosswalk
specificationVersion: '0.1'
provider: AIMLAPI
providerId: aimlapi
generated: '2026-08-30'
method: derived
source: >-
  Capability groups published at https://docs.aimlapi.com/quickstart/mcp bound to
  operationIds read from the provider's own OpenAPI at
  https://api.aimlapi.com/docs-yaml (saved as
  openapi/aimlapi-inference-openapi.yml) and from the per-endpoint OpenAPI
  fragments embedded in the service-endpoint documentation pages.
description: >-
  Crosswalk between the AIMLAPI remote MCP server and the REST operations that
  back it. IMPORTANT PROVENANCE CAVEAT: mcp.aimlapi.com/mcp gates tools/list
  behind authentication, so the LEFT-HAND SIDE OF THIS TABLE IS NOT A TOOL LIST.
  It is the four capability groups the provider itself publishes, bound to the
  REST operations that must be serving them. No tool name here has been observed
  on the wire and no inputSchema has been captured; every confidence below is
  therefore medium at best. Re-run with an authenticated tools/list to replace
  these rows with real tool names and schemas.
surfaces:
  openapi:
    file: openapi/aimlapi-inference-openapi.yml
    source: https://api.aimlapi.com/docs-yaml
    gated: false
    operations: 19
    note: >-
      19 operations across 16 paths. Three paths reuse one operationId for both
      their GET and their POST (_v2_video_generations, _v1_batches,
      _v2_generate_audio), so there are only 16 distinct operationIds.
  service_endpoints:
    source: https://docs.aimlapi.com/api-references/service-endpoints
    gated: false
    note: >-
      Account, key-management, usage, model-catalogue and deprecation operations
      are documented with their own inline OpenAPI 3.0.0 fragments on the docs
      pages but are ABSENT from the /docs-yaml document. They are real, callable
      operations with no entry in the published master spec.
  mcp:
    url: https://mcp.aimlapi.com/mcp
    gated: true
    note: tools/list returns 401 invalid_key without a bearer token or OAuth session.
  graphql:
    url: null
    note: AIMLAPI publishes no GraphQL surface.
crosswalk:
- tool: model discovery
  category: catalogue
  rest:
  - GET /v1/models
  - GET /v1/models/deprecations
  binding: capability-group
  confidence: medium
  note: >-
    Both backing operations are public and unauthenticated; GET /v1/models
    returned 200 with 519,560 bytes and GET /v1/models/deprecations returned 200
    with 132 entries on 2026-08-30. Neither has an entry in the published master
    OpenAPI, so the tool's inputSchema cannot be inherited from it — only from
    the inline fragment on the docs page.
- tool: inference (chat / text)
  category: inference
  rest:
  - _v1_chat_completions
  - _v1_responses
  - _v1_responses_:response_id
  - _v1_messages
  binding: capability-group
  confidence: medium
  note: >-
    Three parallel text surfaces: the OpenAI-compatible chat completions endpoint,
    an OpenAI Responses-shaped endpoint, and an Anthropic Messages-shaped
    endpoint. The MCP server's "LLM chat" capability is served by at least one of
    them; which one is not published.
- tool: inference (embeddings)
  category: inference
  rest:
  - _v1_embeddings
  binding: capability-group
  confidence: medium
- tool: inference (image)
  category: inference
  rest:
  - _v1_images_generations
  - _v1_images_edits
  binding: capability-group
  confidence: medium
- tool: inference (video)
  category: inference
  rest:
  - _v2_video_generations
  binding: capability-group
  confidence: medium
  note: One operationId covers both the POST that submits and the GET that polls.
- tool: inference (audio / speech)
  category: inference
  rest:
  - _v1_tts
  - _v1_stt_create
  - _v1_stt_:generation_id
  - _v2_generate_audio
  - _v2_generate_audio_preprocess
  binding: capability-group
  confidence: medium
- tool: job management
  category: jobs
  rest:
  - _v2_video_generations
  - _v1_stt_:generation_id
  - _v2_generate_audio
  - _v1_batches
  - _v1_batches_cancel_:batch_id
  binding: capability-group
  confidence: medium
  note: >-
    The docs describe "submit long-running generations and poll their
    status/results", which is exactly the submit/poll pair on the video, stt and
    audio endpoints plus the batch surface.
- tool: account (balance and usage)
  category: account
  rest:
  - GET /v2/billing
  - GET /v2/usage
  - _v2_logs
  binding: capability-group
  confidence: medium
  note: >-
    "What's my AIMLAPI balance?" is the provider's own worked example of an MCP
    call, and GET /v2/billing is the documented balance endpoint.
mcp_only: []
rest_only:
- operationId: _v1_ocr
  note: >-
    OCR is a published REST operation with no counterpart in any documented MCP
    capability group.
- operationId: KeysController_createKey_v1
  note: >-
    API key creation, listing, update and deletion (POST/GET /v1/keys,
    PATCH/DELETE /v1/keys/{prefix}, GET /v1/key) require a management key and are
    not described as an MCP capability. Correctly so — a credential-minting
    operation is the wrong thing to expose to an agent.
- operationId: _v1_models_deprecations
  note: >-
    Public deprecation feed. Reachable through "model discovery" only by
    inference; not named as a capability.
coverage:
  mcp_tools_observed: 0
  mcp_capability_groups_published: 4
  rest_operations_in_master_spec: 19
  rest_operations_documented_outside_master_spec: 12
  bound_rows: 8
  unbound_rest: 3
  note: >-
    Coverage cannot be computed honestly against tool names because none were
    observed. The one number that IS solid: 12 documented, callable operations
    (keys, key usage, usage logs, billing, transactions, model catalogue, model
    metrics, deprecations) exist entirely outside the published master OpenAPI.