Web Fetch

import { Aside, Steps } from ‘@astrojs/starlight/components’;

The Web Fetch tool lets the AI assistant fetch and read content from any URL — documentation pages, API specs, articles, JSON endpoints, and more. When you ask the assistant to “read this page” or “look up that docs”, it uses this tool automatically.

What you can do

• Read documentation — point the assistant at any docs page and ask questions about it
• Fetch API specs — paste an OpenAPI or JSON endpoint URL; the assistant reads and explains it
• Research while you work — ask the assistant to compare two libraries by reading their READMEs
• Get page summaries — “summarize https://…” without leaving the chat

Enabling Web Fetch

Web Fetch is enabled by default. An org admin can disable it in the org settings.

1. Go to **Organizations** → open your org → **Tools** tab 2. Find the **AI Assistant Tools** section 3. Toggle **Web Fetch** on

The setting applies to all members of your organization. There is no per-user or per-project control at this time.

Using it in the assistant

Once enabled, just mention a URL in the chat — the assistant will fetch it when relevant:

You:       Read https://docs.example.com/quickstart and tell me how to install it
Assistant: Fetching the page... [fetch_page called]
           Here's how to install: ...

You can also be explicit:

You: Fetch https://api.example.com/openapi.json and list all available endpoints
You: Compare https://lodash.com/docs and https://ramdajs.com — which is better for our use case?

Supported content types

Content type	How it’s handled
HTML page	Extracted with Mozilla Readability — clean text, no nav/ads
JSON endpoint	Pretty-printed JSON
Plain text / Markdown	Returned as-is

Pages that require JavaScript to render (single-page apps, dashboards) may not return useful content — only the initial HTML is fetched, not the rendered DOM.

Limits

Parameter	Value
Timeout	30 seconds
Max response size	2 MB
Max characters returned	50 000 (default), up to 100 000

If the content exceeds the character limit it will be truncated. You’ll see a note in the assistant’s response when this happens.

Tips

Ask for a specific section — instead of fetching a huge docs site, link directly to the page you need.

Use max_chars for long pages — if you need more content, ask the assistant explicitly:

Fetch https://... with max 80000 characters

JSON APIs work great — point the assistant at any JSON API endpoint and it will parse and explain the structure automatically.