Skip to main content
MCP server CX Codex

MCP Retrieval — Claude Code MCP server for web search and scraping

MCP Retrieval — a Go MCP server with three tools: DuckDuckGo search, Bing Images, and page scraping into Markdown. No API keys required.

MIT ★ 23 updated 2026-10-24

Install into Claude Code · Platform CLI All methods and platforms ↓
claude mcp add --env MCP_TRANSPORT=<MCP_TRANSPORT> --env MCP_NAME=<MCP_NAME> --env MCP_PATH=<MCP_PATH> --env SERVER_PORT=<SERVER_PORT> --env SERVER_READ_TIMEOUT=<SERVER_READ_TIMEOUT> --transport stdio mcp-retrieval -- docker run -i --rm -e MCP_TRANSPORT -e MCP_NAME -e MCP_PATH -e SERVER_PORT -e SERVER_READ_TIMEOUT ghcr.io/role1776/mcp-retrieval:0.1.10
Open on GitHub

What MCP Retrieval does

The server hands the model three read-only tools. web_search runs queries against DuckDuckGo Lite, web_search_images pulls pictures from Bing Images, and web_scrape downloads pages and extracts the main article into Markdown. Each tool takes a list of queries or URLs and runs them in parallel. Results come back as structured JSON; clients that ignore structuredContent get the same data mirrored into a text block by the SDK.

Under the hood the retrieval-go library does the fetching. It impersonates real browsers at the TLS layer via uTLS, with optional rotation of browser fingerprints and proxies to stay ahead of bot protection. Two transports are available: stdio by default, plus http for remote deployments. Both expose the same toolset.

You run it as a Docker container — ghcr.io/role1776/mcp-retrieval:0.1.10 — configured with MCP_TRANSPORT, MCP_NAME, MCP_PATH, SERVER_PORT, and SERVER_READ_TIMEOUT. Queries cap at 512 characters, URLs at 2048 and only http/https. Deduplication happens per query, not across the whole call: the same link found by two queries shows up in both results.

What it helps with

  • run web search for an agent without signing up for an API
  • pull clean article text from a URL into the model context
  • fetch images for several queries in one call
  • avoid basic bot detection when hitting search engines
  • deploy the server over HTTP instead of stdio

Example prompts for the agent

  • Search the web for recent pages about MCP servers and return five snippets.
  • Scrape this article and give me the text without links, capped at 10000 characters.
  • Find images for "Claude Code architecture" from the past week.

How to install MCP Retrieval

Platform CLI

claude mcp add --env MCP_TRANSPORT=<MCP_TRANSPORT> --env MCP_NAME=<MCP_NAME> --env MCP_PATH=<MCP_PATH> --env SERVER_PORT=<SERVER_PORT> --env SERVER_READ_TIMEOUT=<SERVER_READ_TIMEOUT> --transport stdio mcp-retrieval -- docker run -i --rm -e MCP_TRANSPORT -e MCP_NAME -e MCP_PATH -e SERVER_PORT -e SERVER_READ_TIMEOUT ghcr.io/role1776/mcp-retrieval:0.1.10

Replace <…> with your own values. 18 more environment variables are in the server README

Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more

Questions about MCP Retrieval

How do I install MCP Retrieval in Claude Code?

Run `claude mcp add --env MCP_TRANSPORT=<MCP_TRANSPORT> --env MCP_NAME=<MCP_NAME> --env MCP_PATH=<MCP_PATH> --env SERVER_PORT=<SERVER_PORT> --env SERVER_READ_TIMEOUT=<SERVER_READ_TIMEOUT> --transport stdio mcp-retrieval -- docker run -i --rm -e MCP_TRANSPORT -e MCP_NAME -e MCP_PATH -e SERVER_PORT -e SERVER_READ_TIMEOUT ghcr.io/role1776/mcp-retrieval:0.1.10`.

What does MCP Retrieval do?

It exposes three tools to the model: DuckDuckGo web search, Bing Images search, and page scraping into Markdown. Each tool processes a list of queries or URLs in parallel.

Do I need API keys?

No. Searches go through DuckDuckGo Lite and Bing Images without keys, and pages are fetched directly. uTLS and browser fingerprint rotation handle bot protection.

Similar MCP servers

All MCP servers for Agent setup: memory, context, orchestration →

Project sponsors

Companies that keep this analytics open