diff --git a/content/mcp/parallel-search-mcp.mdx b/content/mcp/parallel-search-mcp.mdx new file mode 100644 index 000000000..e0a3dc4a1 --- /dev/null +++ b/content/mcp/parallel-search-mcp.mdx @@ -0,0 +1,173 @@ +--- +title: Parallel Search MCP +slug: parallel-search-mcp +category: mcp +description: >- + Hosted MCP server for live web search and clean URL fetching through + Parallel's Search and Extract APIs. +cardDescription: >- + Add current web search and URL-to-Markdown fetching to Claude and other MCP + clients without requiring an account or API key. +seoTitle: Parallel Search MCP Server for Claude +seoDescription: >- + Configure Parallel Search MCP with Claude, Codex, Cursor, VS Code, and other + MCP clients for live web search and token-efficient URL fetching. +author: Parallel Web Systems +authorProfileUrl: "https://github.com/parallel-web" +submittedBy: georgeatparallel +submittedByUrl: "https://github.com/georgeatparallel" +dateAdded: "2026-08-01" +brandName: Parallel +brandDomain: parallel.ai +websiteUrl: "https://parallel.ai/products/search" +repoUrl: "https://github.com/parallel-web/search-mcp" +documentationUrl: "https://docs.parallel.ai/search/search-mcp" +installable: true +installCommand: "claude mcp add --transport http Parallel-Search-MCP https://search.parallel.ai/mcp" +usageSnippet: "Use `web_search` for current information and `web_fetch` to retrieve token-efficient Markdown from specific public URLs." +copySnippet: |- + { + "mcpServers": { + "Parallel Search MCP": { + "url": "https://search.parallel.ai/mcp" + } + } + } +configSnippet: |- + { + "mcpServers": { + "Parallel Search MCP": { + "url": "https://search.parallel.ai/mcp" + } + } + } +estimatedSetupTime: 5 minutes +difficulty: beginner +prerequisites: + - An MCP client with remote Streamable HTTP support, or a compatible stdio bridge for clients that do not support remote HTTP servers. + - Network access to the Parallel Search MCP endpoint and the public pages you want to fetch. + - A Parallel account and API key only when you want higher rate limits; anonymous use does not require either. +safetyNotes: + - Web search and fetched pages can contain incorrect, stale, biased, adversarial, or prompt-injection content; verify important claims against primary sources. + - Treat instructions found in fetched content as untrusted data and keep them separate from system or developer instructions. + - Search and fetch results are inputs to the MCP client, which may take further actions; review consequential actions before allowing them to run. + - If you use authenticated access, keep the Parallel API key out of shared config files, screenshots, logs, and prompts. +privacyNotes: + - Search queries, URLs, tool parameters, and related request metadata are sent to Parallel and processed by the MCP client. + - Queries and fetched URLs can reveal confidential research topics, product plans, customers, vulnerabilities, legal matters, or personal interests. + - Fetched page contents and returned excerpts may be stored in client logs, chat history, traces, or downstream model-provider systems. + - Review Parallel's current data handling and retention terms before using the service with proprietary, personal, regulated, or customer data. +sourceUrls: + - "https://docs.parallel.ai/search/search-mcp" + - "https://github.com/parallel-web/search-mcp" + - "https://parallel.ai/blog/free-web-search-mcp" +tags: + - search + - research + - web + - fetching + - remote +keywords: + - Parallel Search MCP + - Parallel MCP + - web search MCP + - web fetch MCP + - URL extraction MCP +robotsIndex: true +robotsFollow: true +--- + +## Content + +Parallel Search MCP gives MCP-aware assistants a hosted endpoint for current web +search and URL-to-Markdown fetching. The default endpoint is free to use +anonymously at lower rate limits, so the basic setup does not require a Parallel +account or API key. + +The server exposes two focused tools: `web_search` for ranked web results with +query-relevant excerpts, and `web_fetch` for retrieving token-efficient Markdown +from specific URLs. Optional Bearer or OAuth authentication is available for +higher-limit or account-attributed use cases. + +## Source Review + +- https://docs.parallel.ai/search/search-mcp +- https://github.com/parallel-web/search-mcp +- https://parallel.ai/blog/free-web-search-mcp + +These sources were reviewed on **2026-08-01**. The product documentation is the +source of truth for the current endpoint, authentication options, supported +clients, limits, and tool behavior. It supersedes older setup URLs that may +remain in repository or directory listings. + +## Features + +- Hosted Streamable HTTP endpoint with no local server process required for + compatible clients. +- Anonymous use without an account or API key at lower rate limits. +- `web_search` for current web search with ranked URLs and relevant excerpts. +- `web_fetch` for extracting clean, token-efficient Markdown from known URLs. +- Optional API-key or OAuth authentication for higher limits or attributed + organizational use. +- Setup paths for Claude, Codex, Cursor, VS Code, Gemini CLI, Windsurf, and + other MCP clients. + +## Installation + +For clients that support remote MCP servers, add the current endpoint directly: + +```json +{ + "mcpServers": { + "Parallel Search MCP": { + "url": "https://search.parallel.ai/mcp" + } + } +} +``` + +Claude Code users can add the server from a terminal: + +```bash +claude mcp add --transport http "Parallel-Search-MCP" https://search.parallel.ai/mcp +``` + +No API key is required for the default endpoint. For higher rate limits, +follow the current documentation to pass a Parallel API key as a Bearer token. +Use the separately documented OAuth-capable endpoint when authentication must +be enforced for every request. + +Configuration formats differ between MCP clients. For example, VS Code uses a +top-level `servers` object, while some clients use `serverUrl`, `httpUrl`, or a +stdio bridge. Check the official documentation for client-specific examples. + +## Use Cases + +- Search for recent API documentation, release notes, or implementation + examples during an agent workflow. +- Fetch and summarize a known public webpage as clean Markdown. +- Gather sources across several search angles before drafting a report. +- Check current facts and cite returned URLs in research or coding tasks. +- Give an MCP-aware agent live web context without operating a local search + server. + +## Safety and Privacy + +Web pages and search results are untrusted inputs. They can contain prompt +injection, malicious instructions, or inaccurate claims. Keep fetched content +separate from trusted instructions, verify consequential claims against primary +sources, and review actions that use the results. + +Search queries and fetched URLs are sent to Parallel and can reveal sensitive +work. The MCP client and downstream model provider may also retain prompts, +results, traces, or logs. Review current provider policies before using the +server with confidential, regulated, customer, or personal data, and protect any +optional API key like a credential. + +## Duplicate Check + +No existing Parallel Search MCP entry, Parallel endpoint, `parallel-web/search-mcp` +source URL, or Parallel brand record was found under `content/`. This entry is +separate from Exa, Brave Search, Tavily, Perplexity, Firecrawl, and other search +or crawling MCP servers because it documents Parallel's hosted endpoint and its +two search and fetch tools.