1Search the web or analyze web pages using AI. Spawns a sub-agent for research tasks.
 2
 3<when_to_use>
 4Use Agentic Fetch when:
 5- Searching the web for information (omit URL)
 6- Extracting specific information from a webpage
 7- Answering questions about web content
 8- Research requiring multiple pages
 9
10Do NOT use Agentic Fetch when:
11- Need raw content without analysis → use `fetch`
12- Need API JSON responses → use `fetch`
13- Downloading files → use `download`
14- Searching local codebase → use `agent`
15</when_to_use>
16
17<parameters>
18- prompt: What information you want (required)
19- url: Specific URL to analyze (optional - omit to search web)
20</parameters>
21
22<modes>
23**Search mode** (no URL): Agent searches web and follows relevant links
24```
25prompt: "What are the new features in Python 3.12?"
26```
27
28**Analysis mode** (with URL): Agent fetches and analyzes specific page
29```
30url: "https://docs.python.org/3/whatsnew/3.12.html"
31prompt: "Summarize the key changes"
32```
33</modes>
34
35<tips>
36- Be specific about what information you want
37- For research, let the agent search (omit URL)
38- Costs more tokens than `fetch` due to AI processing
39- If MCP web tools available (mcp_*), prefer those
40</tips>
41
42<limits>
43- 5MB per page
44- HTTP/HTTPS only
45- Some sites block automated requests
46- Search depends on DuckDuckGo availability
47</limits>