Source: Claude web fetch tool docs

Anthropic's reference for the server-side web_fetch tool: URL validation against exfiltration, content limits, caching, citations and dynamic filtering.

Contents

Summary

A server tool that retrieves the full text of a web page or PDF so Claude can read it. It pairs with web search: search finds candidate URLs, fetch reads the chosen ones.

Key claims

  • Versions (as of 2026-11): web_fetch_20250910 (basic), 20260209 (dynamic filtering), 20260309 (adds use_cache cache bypass), 20260318 (adds response_inclusion).
  • Exfiltration guard: Claude may only fetch URLs that already appeared in user messages, client tool results or earlier search/fetch results, never URLs it invents; URLs that look like they contain credentials are refused. url_sources narrows the allowed sources further. The docs still warn of residual risk when untrusted input meets sensitive data.
  • Does not render JavaScript; for such pages the docs point to a browser-use tool.
  • max_content_tokens truncates long content; PDFs come back as base64 documents. Rough sizes: average page ~2.5k tokens, large doc page ~25k, research PDF ~125k.
  • Results are cached by default and may be stale; use_cache: false forces a fresh fetch at the cost of latency.
  • Citations are optional (off by default) and use character locations in the fetched document.
  • No per-fetch fee; only tokens are billed.
  • Error codes include url_not_in_prior_context, url_not_allowed (including robots.txt and private addresses), url_not_accessible, unsupported_content_type, content_too_large.

Feeds

Web search APIs, Deep research agents

Suggest changes →