generated: '2026-08-14'
method: searched
source: live MCP tools/list against https://search.parallel.ai/mcp (HTTP 200) + https://docs.parallel.ai/integrations/mcp/search-mcp
status: published
status_raw: live
name: Parallel Web Search MCP Server
version: 1.27.0
website: https://parallel.ai
docs: https://docs.parallel.ai/integrations/mcp/search-mcp
transport: streamable-http
protocol_version: '2025-06-18'
deployment:
  mode: remote
  verified: searched
  tools: 2
  checked: '2026-08-12'
  source: catalog MCP census
endpoints:
- url: https://search.parallel.ai/mcp
  auth: anonymous
  probed_status: 200
  notes: 'Default endpoint. Free anonymous access at lower rate limits; pass a Parallel API key as ''Authorization:
    Bearer <key>'' for higher limits. OAuth is deliberately NOT advertised here, so OAuth-capable clients
    will not prompt for sign-in.'
- url: https://search.parallel.ai/mcp-oauth
  auth: oauth
  notes: OAuth-capable variant. Requires authentication - Bearer API key OR the OAuth flow; anonymous
    requests return 401. Recommended for org-wide deployments, Zero Data Retention setups, or wherever
    every request must be attributed to an account.
capabilities:
  tools: true
  prompts: true
  resources: true
tool_count: 2
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. Ideal for questions needing fresh or current
    information, research/comparison/documentation/troubleshooting, and broad tasks where multiple search_queries
    can be issued in a single call.
  backing_rest: POST /v1/search
  input_schema:
    required:
    - objective
    - search_queries
    properties:
    - name: objective
      type: string
      detail: Natural-language description of what the search is trying to find. Should be atomic; may
        include guidance about preferred sources or freshness.
    - name: search_queries
      type: array[string]
      detail: Concise keyword queries, 3-6 words each, may include search operators. At least one required;
        2-3 recommended.
    - name: session_id
      type: string
      detail: Stable conversation identifier (UUID or 32+ char hex) reused across every web_search/web_fetch
        call. Used for free-tier rate limiting and log correlation; ignored on paid-tier keys. maxLength
        100.
    - name: model_name
      type: string
      detail: Identifier of the calling LLM model, passed verbatim. Product analytics only; does not affect
        search behavior. maxLength 100.
  output_schema: true
  output_detail: Structured results (url, title, publish_date, excerpts) plus a usage[] block of SKU name/count
    items - the MCP response carries its own billing telemetry.
- name: web_fetch
  description: Pulls token-efficient markdown from specific URLs. Use after web_search narrows down candidates,
    or when the agent already has a URL it needs to read in depth. Accepts multiple related URLs plus
    a common objective per call.
  backing_rest: POST /v1/extract
  input_schema:
    required:
    - urls
    properties:
    - name: urls
      type: array[string]
    - name: objective
      type: string
    - name: search_queries
      type: array[string]
    - name: full_content
      type: boolean
    - name: session_id
      type: string
    - name: model_name
      type: string
  output_schema: true
constraints:
  search_mode: The MCP wrapper runs Search in `basic` mode - tuned for low latency inside agent loops.
  excerpt_cap: Total excerpts per tool call are capped at roughly 25,000 characters to stay within typical
    MCP client output limits.
install_clients:
  one_click:
  - Cursor
  - VS Code
  - LM Studio
  - Goose
  config_file:
  - Claude Code
  - Codex CLI
  - Claude Desktop
  - Windsurf
  - Zed
  - Gemini CLI
  - Warp
  - Kiro
  programmatic: https://docs.parallel.ai/integrations/mcp/programmatic-use
crosswalk: mcp/parallel-web-systems-tool-crosswalk.yml
related:
  task_mcp_repo: https://github.com/parallel-web/task-mcp
  task_mcp_note: A task-mcp repo exists under the GitHub org but no hosted endpoint or published package
    for it was found; it is not counted as a deployed server here.
notes: Upgraded from a bare initialize handshake to a full live tools/list - both tools were returned
  anonymously with complete inputSchema AND outputSchema, so the tool contract in this file is the server's
  own, not doc prose. Use is subject to the Parallel Customer Terms and Privacy Policy.