Web Research MCP

MCP server for web research with DuckDuckGo search and content extraction.

No API keys. No configuration. Just npx and go.

Table of Contents

• Quick Start
• What It Does
• Looking for Pre-Built Agents?
• Tools
  • multi_search
  • fetch_pages
• How It Works
• License

Quick Start

Add the server to your MCP client config. No install needed — npx handles everything.

Claude Code (.mcp.json in project root):

{
  "mcpServers": {
    "web-research": {
      "command": "npx",
      "args": ["web-research-mcp"]
    }
  }
}

OpenCode (opencode.json):

{
  "mcpServers": {
    "web-research": {
      "command": "npx",
      "args": ["web-research-mcp"]
    }
  }
}

Or run it directly:

npx web-research-mcp

What It Does

Two MCP tools for AI assistants to search the web and extract page content:

• multi_search — Search DuckDuckGo with multiple queries in parallel
• fetch_pages — Fetch and extract clean text from web pages

No API keys required. Works with any MCP-compatible client.

Looking for Pre-Built Agents?

If you want ready-to-use AI agents for web research (not just the raw tools), check out web-research-toolkit. It installs pre-built agents and skills for Claude Code and OpenCode — one command, fully configured.

npx web-research-toolkit install opencode

Tools

multi_search

Search the web using multiple queries via DuckDuckGo. Returns deduplicated results across all queries.

| Parameter | Type | Required | Default | Description | |-----------|------|----------|---------|-------------| | queries | string[] | Yes | — | Search queries to execute | | results_per_query | number | No | 5 | Number of results per query |

Returns: Deduplicated URLs, snippets, titles, and per-query mapping.

Example call:

{
  "queries": ["rust async runtime", "tokio vs async-std"],
  "results_per_query": 3
}

fetch_pages

Fetch and extract clean text content from multiple web pages in parallel.

| Parameter | Type | Required | Default | Description | |-----------|------|----------|---------|-------------| | urls | string[] | Yes | — | URLs to fetch | | max_chars | number | No | 15000 | Max characters per page | | timeout | number | No | 30 | Timeout in seconds per request |

Returns: Extracted text content, page titles, and per-URL error details.

Example call:

{
  "urls": ["https://docs.rs/tokio/latest/tokio/", "https://async.rs/"],
  "max_chars": 10000
}

How It Works

• Search — Scrapes DuckDuckGo HTML results (no API key needed). Runs queries in parallel and deduplicates URLs across results.
• Content extraction — Uses Mozilla Readability to extract article text from raw HTML, stripping navigation, ads, and boilerplate.
• Smart truncation — Limits extracted text to max_chars to keep responses within context window limits.
• Parallel fetching — Fetches multiple pages concurrently with per-request timeouts.

License

MIT