Web Research MCP — Search, Fetch & Synthesize

An MCP server for AI agents that need to search the web, read pages, and synthesize research — all in a single tool call.

Pay-per-use. No subscription. No idle charges. Powered by Apify Standby.

---

What it does

Three tools, designed to work together:

Tool	What it does
web_search	Search the web. Returns titles, URLs, and snippets. Uses Brave Search API (if configured) or DuckDuckGo fallback.
fetch_page	Fetch any URL and return clean markdown. Strips ads, nav, and boilerplate. Supports article, full-body, or links-only extraction.
research_topic	Multi-step research in one call: search → fetch → score relevance → deduplicate → extract key facts → synthesize.

---

Quick connect

Add to your MCP client config (Claude Desktop, Cursor, etc.):

{
  "mcpServers": {
    "web-research": {
      "type": "http",
      "url": "https://idapixl--web-research-mcp.apify.actor/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_APIFY_API_TOKEN"
      }
    }
  }
}

Replace YOUR_APIFY_API_TOKEN with your token from apify.com/account/integrations.

Claude Code:

claude mcp add web-research https://idapixl--web-research-mcp.apify.actor/mcp --transport http

Cursor / Windsurf — add via Settings > MCP Servers > HTTP URL.

---

Pricing

Pay only for what you use. Charged per tool call.

Event	Price
web_search	$0.003
fetch_page	$0.005
research_topic	$0.025 (single charge, includes all sub-calls)

---

Brave Search API setup (optional but recommended)

By default the actor uses DuckDuckGo HTML scraping, which works without any key. For better results and freshness filtering:

1. Go to brave.com/search/api and sign up for the free tier (2,000 queries/month)
2. Copy your API key
3. In the actor input, set Brave Search API Key (stored as secret — never exposed in logs)

When the key is set, web_search and research_topic will use Brave Search automatically. DuckDuckGo is the fallback on error.

---

Example tool calls

Search with freshness filter (Brave only)

{
  "tool": "web_search",
  "arguments": {
    "query": "Claude 4 release announcement",
    "maxResults": 5,
    "freshness": "pw"
  }
}

Read a page — links extraction mode

{
  "tool": "fetch_page",
  "arguments": {
    "url": "https://docs.anthropic.com/en/api/getting-started",
    "extractMode": "links"
  }
}

Deep research with focus

{
  "tool": "research_topic",
  "arguments": {
    "topic": "Model Context Protocol",
    "maxSources": 3,
    "focus": "pricing and monetization"
  }
}

---

Input configuration

Field	Type	Default	Description
braveApiKey	string (secret)	—	Brave Search API key for higher-quality search
maxSearchResults	integer	5	Max results per web_search call (1–10)
fetchTimeoutSecs	integer	30	Page fetch timeout in seconds (5–120)
proxyConfiguration	object	—	Apify Proxy config (residential recommended for best success rates)

---

Tool reference

web_search

• query (required) — search query, max 500 chars
• maxResults (optional) — 1–10, default 5
• freshness (optional) — pd past day, pw past week, pm past month, py past year (Brave only)

fetch_page

• url (required) — page URL
• timeoutSecs (optional) — 5–120, default 30
• extractMode (optional) — article main content (default), full entire body, links all links

research_topic

• topic (required) — research topic or question, max 500 chars
• maxSources (optional) — 1–5, default 3
• focus (optional) — sub-question to narrow the research (e.g. "performance benchmarks"), max 200 chars

---

Notes

• Stateless: each MCP request gets an isolated server instance
• research_topic scores relevance, deduplicates near-duplicate sources (>70% sentence overlap filtered), and extracts key sentences containing your query terms
• Proxy: Apify Residential proxies improve success rates on pages that block datacenter IPs
• Charges fire before execution — if a page fails to load after charging, the tool returns an error

---

Built by Idapixl. TypeScript, MCP SDK 1.x, Apify SDK 3.x.