Agents - New web fetch tool in the Agents SDK for reading webpages and documents, with automatic markdown conversion
Cloudflare Agents SDK introduces a web fetch tool that retrieves webpages, PDFs, and Office documents, converting them into Markdown for LLMs.
Why this signal matters
The Cloudflare Agents SDK (v0.28.0 or later) now includes a web fetch tool capable of ingesting webpages, PDFs, and supported Office documents. It utilizes Workers AI Markdown Conversion to translate content into text for language models, requesting Markdown directly when supported by sites or formatting JSON responses. The tool returns page content, the resolved URL, HTTP status code, and content type. Long content is returned in paginated sections with offsets for incremental reading. Operators can configure allowed host lists, maximum response sizes, timeouts, and redirect handling. It also integrates with the SDK's web search tool, prompting models to follow up searches by fetching returned pages.
Actionable summary
Cloudflare Agents SDK v0.28.0+ adds webFetchTool ( agents/webfetch/pi ), leveraging Workers AI Markdown Conversion to fetch and convert web pages, PDFs, and Office files into Markdown/JSON with chunking support.
- Agent usefulness
- 80/100
- Confidence
- 90%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade to Cloudflare Agents SDK v0.28.0 or higher.
- Import and configure webFetchTool from agents/webfetch/pi in the agent harness.
- Provide a Workers AI binding ( this.env.AI ) to the webFetchTool configuration.
- Optionally configure host restrictions, timeouts, response sizes, and redirect rules.