---

name: exa-fetch description: Fetch and read the full content of one or more URLs using Exa via the mcp__claude_ai_exa__web_fetch_exa tool. Use when the user provides a URL to read, asks to summarize or extract information from a specific webpage, or when exa-search returns results that need deeper reading. Batches multiple URLs in a single call.

Exa Web Fetch

Retrieve full page content from known URLs as clean markdown. Batches multiple URLs in one call.

Instructions

1. Determine maxCharacters

| Use case | maxCharacters | |---|---| | Quick summary / key facts | 2000–3000 | | Normal reading / analysis | 5000–8000 | | Full document / deep research | 15000–20000 |

Default: 3000 (Exa default). Increase only when the task requires completeness.

2. Batch when possible

Always batch multiple URLs into a single call — one call, multiple URLs:

mcp__claude_ai_exa__web_fetch_exa(
  urls=["https://example.com/a", "https://example.com/b"],
  maxCharacters=5000
)

3. Call the tool

mcp__claude_ai_exa__web_fetch_exa(
  urls=["<url1>", "<url2>", ...],
  maxCharacters=<n>
)

4. Handle the response

• Content is returned as clean markdown — use it directly
• If content is truncated and more is needed, re-fetch with a higher maxCharacters
• If a page returns empty/minimal content, note it to the user (may be paywalled or JS-rendered)

Examples

User provides a URL to read:

mcp__claude_ai_exa__web_fetch_exa(
  urls=["https://example.com/article"],
  maxCharacters=5000
)

Follow-up from exa-search — reading top 3 results:

mcp__claude_ai_exa__web_fetch_exa(
  urls=[
    "https://site1.com/post",
    "https://site2.com/page",
    "https://site3.com/doc"
  ],
  maxCharacters=4000
)

User asks to compare two documentation pages:

mcp__claude_ai_exa__web_fetch_exa(
  urls=["https://docs.framework-a.dev/routing", "https://docs.framework-b.dev/routing"],
  maxCharacters=8000
)

Tips

• Prefer fetch over search when the URL is already known — it's faster and more targeted
• For API docs or reference pages, 8000–12000 characters usually captures full relevant content
• Batch aggressively: 5–10 URLs in one call is fine and saves round-trips
• If a page appears empty, mention to the user that the content may require JavaScript or authentication