Web Fetch

Not yet available. The Web Fetch server tool is not yet available.

Web Fetch retrieves the text content of a URL and injects it into the conversation before the model generates a reply. It handles HTML stripping, Markdown extraction, and content truncation so the model receives clean, token-efficient input rather than raw markup.

Enabling web fetch#

{
  "model": "anthropic/claude-opus-4.8",
  "messages": [
    {
      "role": "user",
      "content": "Summarise the content at https://example.com/article"
    }
  ],
  "tools": [
    { "type": "opendunes.web_fetch" }
  ]
}

The model determines which URL to fetch based on the conversation. It can also choose not to fetch if no URL is relevant.

Explicit URL hint#

You can instruct the gateway to pre-fetch a specific URL regardless of whether the model requests it:

{
  "type": "opendunes.web_fetch",
  "options": {
    "url": "https://docs.example.com/reference",
    "always_fetch": true
  }
}

With always_fetch: true, the content is injected at the start of the context before the model's first token is generated.

Configuration#

Content extraction#

The gateway strips navigation, ads, and boilerplate using an HTML-to-Markdown extractor. The result is a compact, readable representation of the page's primary content. Use "raw" only when you need exact HTML (counts heavily against your token budget).

Allowed domains#

By default any HTTPS URL may be fetched. You can restrict fetch targets to an allowlist in your workspace settings to prevent prompt-injection via attacker-controlled URLs.

Billing#

Web Fetch charges a flat fee per URL fetched, plus the token cost of the injected content. Very large pages are truncated before billing, so max_chars is an effective cost control.