generated: '2026-08-14'
method: derived
source: >-
  mcp/parallel-mcp.yml (live tools/list on search.parallel.ai + Task MCP docs)
  bound to openapi/_original/parallel-openapi-original.json operationIds
notes: >-
  Binds every published Parallel MCP tool to the REST operation(s) it fronts. The
  two Search MCP tools were introspected live, so their real inputSchema is known
  and the binding is high confidence. The four Task MCP tools are auth-gated;
  they are mapped from the provider's own tool descriptions, which name the
  backing product surface explicitly, so confidence is high on the create/status
  pair and medium where a tool spans a family of operations.
surfaces:
  openapi: openapi/_original/parallel-openapi-original.json
  openapi_public: https://docs.parallel.ai/docs-latest-openapi.json
  graphql: null
  mcp:
  - {url: 'https://search.parallel.ai/mcp', gated: false, introspected: true}
  - {url: 'https://task-mcp.parallel.ai/mcp', gated: true, introspected: false}
  a2a: {url: 'https://api.parallel.ai/a2a', gated: true}
crosswalk:
- tool: web_search
  server: parallel-search
  category: search
  rest: [v1_search_v1_search_post]
  binding: one-to-one
  confidence: high
  note: >-
    Live inputSchema (objective, search_queries) mirrors the V1SearchRequest body
    of POST /v1/search. The MCP tool pins `basic` mode and caps excerpt volume.
- tool: web_fetch
  server: parallel-search
  category: extraction
  rest: [extract_v1_extract_post]
  binding: one-to-one
  confidence: high
  note: >-
    Live inputSchema (urls, objective, search_queries, full_content) mirrors the
    V1ExtractRequest body of POST /v1/extract.
- tool: createDeepResearch
  server: parallel-task
  category: research
  rest: [tasks_runs_post_v1_tasks_runs_post]
  binding: one-to-one
  confidence: high
  note: Provider docs bind this tool to the Task API run-creation surface.
- tool: createTaskGroup
  server: parallel-task
  category: research-batch
  rest:
  - tasks_taskgroups_post_v1_tasks_groups_post
  - tasks_taskgroups_runs_post_v1_tasks_groups__taskgroup_id__runs_post
  binding: composite
  confidence: high
  note: >-
    Creating a group and adding runs to it are two REST operations; the tool
    presents them as one enrichment action.
- tool: getStatus
  server: parallel-task
  category: status
  rest:
  - tasks_runs_get_v1_tasks_runs__run_id__get
  - tasks_taskgroups_get_v1_tasks_groups__taskgroup_id__get
  binding: one-to-many
  confidence: medium
  note: >-
    Docs describe a lightweight (~50 token) status check; it resolves to the run
    or the group status endpoint depending on the id supplied.
- tool: getResultMarkdown
  server: parallel-task
  category: results
  rest:
  - tasks_runs_result_get_v1_tasks_runs__run_id__result_get
  - tasks_taskgroups_runs_get_v1_tasks_groups__taskgroup_id__runs_get
  binding: one-to-many
  confidence: medium
  note: >-
    REST returns structured JSON output; the tool returns an LLM-friendly markdown
    rendering, so this is a transform over the result operations rather than a
    pass-through.
mcp_only: []
a2a_skills:
- skill: research
  agent: https://api.parallel.ai/a2a
  rest: [tasks_runs_post_v1_tasks_runs_post, tasks_runs_result_get_v1_tasks_runs__run_id__result_get]
  binding: composite
  confidence: medium
  note: >-
    The A2A "Deep Research" skill returns a cited markdown report and supports
    follow-ups in the same context; the card's documentationUrl points at the Task
    API quickstart.
rest_only:
- tasks_sessions_events_get_v1_tasks_groups__taskgroup_id__events_get
- tasks_taskgroups_runs_id_get_v1_tasks_groups__taskgroup_id__runs__run_id__get
- tasks_runs_events_get_v1_tasks_runs__run_id__events_get
- tasks_runs_input_get_v1_tasks_runs__run_id__input_get
- tasks_runs_events_get_v1beta_tasks_runs__run_id__events_get
- findall_entity_search_v1beta_findall_entity_search_post
- ingest_findall_run_v1beta_findall_ingest_post
- findall_runs_v1_v1beta_findall_runs_post
- findall_runs_v1_get_v1beta_findall_runs__findall_id__get
- cancel_findall_run_v1beta_findall_runs__findall_id__cancel_post
- enrich_findall_run_v1beta_findall_runs__findall_id__enrich_post
- get_findall_events_v1beta_findall_runs__findall_id__events_get
- extend_findall_run_v1beta_findall_runs__findall_id__extend_post
- get_findall_result_v1beta_findall_runs__findall_id__result_get
- get_findall_schema_v1beta_findall_runs__findall_id__schema_get
- create_monitor_v1_monitors_post
- list_monitors_v1_monitors_get
- retrieve_monitor_v1_monitors__monitor_id__get
- cancel_monitor_v1_monitors__monitor_id__cancel_post
- list_monitor_events_v1_monitors__monitor_id__events_get
- trigger_monitor_run_v1_monitors__monitor_id__trigger_post
- update_monitor_v1_monitors__monitor_id__update_post
- clear_memory_v1beta_memory_clear_post
- evict_memory_source_v1beta_memory_evict_post
- retrieve_memory_v1beta_memory_retrieve_post
- chat_completions_v1beta_chat_completions_post
- create_response_v1_responses_post
coverage:
  rest_operations_total: 34
  rest_operations_mapped: 8
  rest_operations_unmapped: 27
  mcp_tools_total: 6
  mcp_tools_mapped: 6
  mcp_tools_introspected: 2
  a2a_skills_total: 1
  note: >-
    Counts overlap by one: tasks_runs_result_get is reachable from both a Task MCP
    tool and the A2A research skill. FindAll, Monitor, Memory, Chat and Responses
    have no MCP tool at all — they are REST/CLI/SDK-only surfaces today.