Web fetch

Fetches an HTTP(S) URL and returns its content. Two modes are available, configured at construction time via make_web_fetch / makeWebFetch:

• agentic (default): HTML is converted to markdown and passed to an analyst agent that answers a prompt, so the full page never enters the main agent’s context. Use when targeted answers are needed about potentially large pages.
• markdown: HTML is converted to clean markdown with scripts, styles, and noise stripped. Use when the agent needs full pages for reasoning.

The max_bytes / maxBytes parameter caps the HTTP response size (default 5 MiB); max_content_chars / maxContentChars caps the extracted content delivered to the model or analyst (default 50,000 characters). For mode='agentic' / mode: 'agentic', the factory also accepts a model for the analyst; agent.aux_modelagent.auxModel is used when none is supplied.

The Python tool delegates all networking to an httpx.AsyncClient. Use the make_web_fetch factory to supply a pre-configured client with custom timeouts, redirects, proxies, or caching.

Supported in: Node.js (TypeScript); Python (all platforms).

Install required

Python

web_fetch requires the optional web-fetch extra:

pip install 'strands-agents[web-fetch]'

TypeScript

HTML conversion requires the optional turndown peer dependency:

npm install turndown

Security posture

Web fetch accepts only http:// and https:// URLs and caps response bodies at 5 MiB by default. Egress control belongs at the layer that encapsulates the agent (a sandbox, microVM, or network policy) where it can be enforced consistently across all tools, including shell.

Example:

TypeScript

import { Agent } from '@strands-agents/sdk'
import { webFetch } from '@strands-agents/sdk/vended-tools/web-fetch'

const agent = new Agent({ tools: [webFetch] })
await agent.invoke('Summarize https://example.com/blog/post')

Reading the full page as markdown:

import { Agent } from '@strands-agents/sdk'
import { makeWebFetch } from '@strands-agents/sdk/vended-tools/web-fetch'

const webFetch = makeWebFetch({ mode: 'markdown' })
const agent = new Agent({ tools: [webFetch] })
await agent.invoke('Read https://example.com/docs and explain the architecture')

Tighter response cap with a dedicated analyst model:

import { Agent } from '@strands-agents/sdk'
import { makeWebFetch } from '@strands-agents/sdk/vended-tools/web-fetch'
import { BedrockModel } from '@strands-agents/sdk/models/bedrock'

const webFetch = makeWebFetch({
  mode: 'agentic',
  maxBytes: 1 * 1024 * 1024,
  maxContentChars: 25_000,
  model: new BedrockModel({ modelId: 'us.amazon.nova-micro-v1:0' }),
})
const agent = new Agent({ tools: [webFetch] })

Python

from strands import Agent
from strands.vended_tools import web_fetch

agent = Agent(tools=[web_fetch])
agent("What is the pricing for the enterprise plan at https://example.com/pricing")

Reading the full page as markdown:

from strands import Agent
from strands.vended_tools import make_web_fetch

web_fetch = make_web_fetch(mode="markdown")
agent = Agent(tools=[web_fetch])
agent("Read https://example.com/docs and then explain the architecture")

Tighter response cap with a dedicated analyst model and custom transport:

import httpx
from strands import Agent
from strands.models import BedrockModel
from strands.vended_tools import make_web_fetch

web_fetch = make_web_fetch(
    mode="agentic",
    client=httpx.AsyncClient(timeout=10.0),
    max_bytes=1 * 1024 * 1024,
    max_content_chars=25_000,
    model=BedrockModel(model_id="us.amazon.nova-micro-v1:0"),
)
agent = Agent(tools=[web_fetch])

📖 Full API Reference

Tags

tool-authoring26