generated: '2026-08-14'
method: probed
status: published
source: >-
  Live MCP tools/list against https://search.parallel.ai/mcp (HTTP 200, real
  inputSchema returned) plus https://docs.parallel.ai/integrations/mcp/quickstart
notes: >-
  CORRECTION vs the 2026-07-20 round: the Search MCP tools are named `web_search`
  and `web_fetch`, not `search`/`extract`. Names and input schemas below are taken
  verbatim from a live anonymous tools/list call. The Task MCP is auth-gated —
  tools/list returns {"error":"Authentication required"} — so its four tools are
  recorded from the provider's own docs (names + descriptions only); their input
  schemas need authenticated introspection.
servers:
- name: parallel-search
  transport: streamable-http
  url: https://search.parallel.ai/mcp
  auth: none
  probe:
    method: tools/list
    http_status: 200
    tools_returned: 2
    checked: '2026-08-14'
  notes: >-
    Free with no account or API key required. Search runs in `basic` mode and
    excerpts per call are capped near 25,000 characters. An optional Bearer API
    key raises rate limits. An OAuth-capable variant is documented at
    https://search.parallel.ai/mcp-oauth.
  tools:
  - name: web_search
    description: >-
      Perform web searches and return LLM-friendly results, including excerpts
      that are usually sufficient to answer directly without a follow-up fetch.
    input_schema:
      required: [objective, search_queries]
      properties:
      - {name: objective, type: string, description: 'Natural-language description of what the web search is trying to find.'}
      - {name: search_queries, type: array, description: 'Concise keyword search queries, 3-6 words each; at least one required, 2-3 recommended.'}
      - {name: session_id, type: string, description: 'Stable conversation identifier; used for free-tier rate limiting.'}
      - {name: model_name, type: string, description: 'Identifier of the calling LLM model; product analytics only.'}
    rest: [v1_search_v1_search_post]
  - name: web_fetch
    description: >-
      Fetch and extract relevant content from specific web URLs. Use only when
      web_search excerpts are insufficient.
    input_schema:
      required: [urls]
      properties:
      - {name: urls, type: array}
      - {name: objective, type: string}
      - {name: search_queries, type: array}
      - {name: full_content, type: boolean}
      - {name: session_id, type: string}
      - {name: model_name, type: string}
    rest: [extract_v1_extract_post]
- name: parallel-task
  transport: streamable-http
  url: https://task-mcp.parallel.ai/mcp
  auth: oauth
  probe:
    method: tools/list
    http_status: 200
    body: '{"error":"Authentication required","detail":"Missing x-api-key or Authorization header..."}'
    checked: '2026-08-14'
    interpretation: Live but auth-gated; tool schemas require authenticated introspection.
  oauth:
    protected_resource_metadata: https://task-mcp.parallel.ai/.well-known/oauth-protected-resource
    authorization_server: https://platform.parallel.ai
    scopes: [key:read]
    see: scopes/parallel-scopes.yml
  notes: >-
    Requires OAuth (dynamic client registration + PKCE) or a Parallel API key as a
    Bearer token from https://platform.parallel.ai. Async architecture: start a
    task, poll status, fetch the result in a later turn.
  tools:
  - name: createDeepResearch
    description: Initiates a deep research task; multi-step research with citations.
    source: https://docs.parallel.ai/integrations/mcp/task-mcp
    rest: [tasks_runs_post_v1_tasks_runs_post]
  - name: createTaskGroup
    description: Initiates a task group to enrich multiple items in parallel.
    source: https://docs.parallel.ai/integrations/mcp/task-mcp
    rest: [tasks_taskgroups_post_v1_tasks_groups_post]
  - name: getStatus
    description: Lightweight status check for an in-flight task (~50 tokens).
    source: https://docs.parallel.ai/integrations/mcp/task-mcp
    rest: [tasks_runs_get_v1_tasks_runs__run_id__get]
  - name: getResultMarkdown
    description: Retrieves the final output of a deep research or task group as markdown.
    source: https://docs.parallel.ai/integrations/mcp/task-mcp
    rest: [tasks_runs_result_get_v1_tasks_runs__run_id__result_get]
deployment:
  mode: remote
  endpoint: https://search.parallel.ai/mcp
  auth: none
  verified: probed
  checked: '2026-08-14'
  additional_endpoints:
  - {url: 'https://task-mcp.parallel.ai/mcp', auth: oauth}
  - {url: 'https://search.parallel.ai/mcp-oauth', auth: oauth}
  install: null
  package: null
  note: >-
    Remote-only. Parallel ships hosted HTTP MCP endpoints; there is no first-party
    stdio package to install. One-click install links exist for Cursor, VS Code,
    LM Studio and Goose, but they configure the same remote URL.
mcp_as_client:
  note: >-
    Parallel is also an MCP CONSUMER — the Task API can call external MCP servers
    as tools during a run.
  docs: https://docs.parallel.ai/task-api/mcp-tool-call
crosswalk: mcp/parallel-tool-crosswalk.yml