Skip to directory

About Parallel Search MCP

Parallel's hosted MCP adds web search and URL content extraction to AI clients, with anonymous Fast mode and account-backed settings for higher limits.

What you can use it for

  • Research current information and verify facts
  • Fetch and summarize specific web pages
  • Gather competitive or market research
Connection and setup
Hosting
Remote hosted
Transport
Streamable HTTP
Endpoint
https://search.parallel.ai/mcp
Authentication
Anonymous free tier or Bearer API key; OAuth-capable /mcp-oauth endpoint also available.

Requirements

  • MCP client supporting remote Streamable HTTP
  • Parallel API account/key or OAuth only for authenticated higher-limit features

Setup steps

  1. Add https://search.parallel.ai/mcp as a remote Streamable HTTP server for anonymous access.
  2. For authenticated use, use a Parallel API key in the Authorization: Bearer header or select the /mcp-oauth endpoint and complete browser sign-in.
  3. For Codex, use codex mcp add parallel-search --url https://search.parallel.ai/mcp; use /mcp-oauth if choosing OAuth.
  4. Use web_search with a focused objective and 2?3 short queries; use web_fetch to read selected URLs.
Tools 2

Legacy standalone Parallel Search MCP exposes exactly web_search and web_fetch and remains available to existing users. For new installs Parallel recommends its newer MCP server, which also exposes web_research and web_decisions.

  • web_search

    Search the web with an objective and concise search queries; returns ranked sources and excerpts.

    01
  • web_fetch

    Extract content from up to 20 supplied URLs as focused excerpts or full markdown.

    02
Limitations and notes
  • Anonymous access is rate-limited and uses server-managed Fast mode.
  • Returned excerpts are capped at roughly 25,000 characters per call.
  • Full page extraction may exceed client output limits.
Original sources
  • official tool reference

    Legacy standalone Search MCP remains available and exposes exactly web_search and web_fetch. New installs should use the newer Parallel MCP server with additional tools.

  • official documentation

    Legacy standalone Search MCP remains available and exposes exactly web_search and web_fetch. New installs should use the newer Parallel MCP server with additional tools.

Documentation checked 2026-10-28T13:02:00.000Z. Connection has not been runtime tested.

Similar tools