web-search-task

name	web_search_task
description	Web search and content retrieval workflow
server_tools	mcp_web__search, mcp_web__fetch, mcp_filesystem__read
client_tools	null
depends_on	null

You have mcp_web__search, mcp_web__fetch and mcp_filesystem__read tools for web research.

Workflow — always follow this sequence:

1. mcp_web__search(query="...") — returns snippets and URLs. Search snippets are summaries, not full content. Never answer based on snippets alone.

2. mcp_web__fetch(url="...") — fetches the page and saves full content to a session_file. Call fetch on the top 1–2 URLs from search results.

3. mcp_filesystem__read(path="<session_file>") — read the saved content. The fetch result includes a session_file path. Use read to access the actual content.

Tool reference

mcp_web__search(query, max_results?, search_depth?, country?) Search the web and return snippets with URLs.

• query: search query string (required)
• max_results: number of results, 1–20 (default: 1). Increase for broad research.
• search_depth: "basic" (default) or "advanced" for deeper search.
• country: ISO code (e.g. "KR") or country name for geo-relevant results.

mcp_web__fetch(url, max_length?, start_index?, extract_mode?) Fetch a URL and save full content to a session_file. Returns a short snippet inline.

• url: the URL to fetch (required)
• max_length: max characters to extract (default: 5000).
• start_index: character offset to start reading from (default: 0). Use for pagination on long pages.
• extract_mode: "markdown" (default) or "text".

mcp_filesystem__read(path, offset?, limit?) Read the saved session_file content.

• path: file path (required)
• offset: start line, 1-indexed.
• limit: max lines to read.

Key rules

• After search, you must call mcp_web__fetch before answering. Search snippets are not sufficient.
• After fetch, you must call mcp_filesystem__read on the session_file. Fetch saves content to a file and does not return it inline.
• For broad topics, run 2–3 mcp_web__search calls in parallel with diverse keywords.
• Call multiple mcp_web__fetch in parallel on different URLs in a single turn.
• After fetches complete, call multiple mcp_filesystem__read in parallel in a single turn.