Web Search MCP Server

What is a Web Search MCP?

A Web Search MCP server exposes web search capabilities as an MCP tool. Instead of writing HTTP requests, parsing JSON and managing API keys in headers, your agent simply calls a tool like searchpipe_search through the MCP protocol. The server handles retrieval, extraction, reranking and returns structured results.

This matters because:

• Zero integration code: MCP clients (Claude Code, Cursor, Codex, OpenCode) handle the protocol. You just provide a URL.
• Standardized interface: The same tool works across every MCP-compatible client. Switch agents without rewriting anything.
• Built-in pipeline: Unlike raw search APIs, a good Web Search MCP bundles extraction and reranking — your agent gets clean, scored results, not raw HTML.

Why agents need web search via MCP

AI agents that can't access the live web are limited to their training data. For research, coding support, fact-checking or any time-sensitive task, real-time web access is essential. MCP makes this trivial:

• Research agents explore topics, gather sources and cite live URLs.
• Coding agents look up documentation, library versions and error solutions on the fly.
• RAG pipelines augment context with fresh web content instead of stale embeddings.
• Autonomous agents verify facts, check prices, monitor news and make decisions grounded in current information.

How SearchPipe implements it

SearchPipe provides a remote MCP server using the streamable-http transport. Key design choices:

• Remote, not local: No npm/pip packages to install, no processes to manage. The server runs on our infrastructure; your agent just needs a URL.
• Key-in-URL authentication: The API key is embedded in the MCP URL (https://searchpipe.tech/mcp/?api_key=sp-YOUR_KEY). This matches the pattern used by Tavily and works with clients that don't support custom headers.
• Single tool, full pipeline: The searchpipe_search tool runs multi-engine retrieval, full-text extraction, LLM reranking and optional answer generation in one call.
• Same pipeline as REST: MCP and the REST API (POST /search) share identical parameters, results and billing. Switching between them costs nothing.

Example: connect in under a minute

For Claude Code:

claude mcp add --transport http searchpipe \
  "https://searchpipe.tech/mcp/?api_key=sp-YOUR_API_KEY"

For other clients, see the per-agent setup guide with exact config snippets for Cursor, Codex, OpenCode, Pi and custom MCP agents.

Your agent can also self-onboard by reading /agent-setup/SKILL.md — a one-page doc that tells the agent how to configure itself.

Supported clients

Any MCP-compatible client that supports streamable-http transport works. Tested and documented:

• Claude Code
• Cursor
• OpenAI Codex
• OpenCode
• Pi coding agent
• Custom MCP agents (any language with an MCP SDK)

Alternatives and trade-offs

• Local stdio MCP servers: Run on your machine. Full control but you manage dependencies, updates and processes. Brave's MCP server is an example.
• REST API only: More flexible for custom integrations, but requires writing HTTP client code. SearchPipe offers both paths equally.
• Browser automation: Playwright/Puppeteer-based approaches. Powerful but slow, fragile and expensive compared to a dedicated search MCP.

SearchPipe's remote MCP server optimizes for zero setup and production reliability: automatic refunds on failure, rate limiting, content moderation and result caching are all handled server-side.