Skip to main content
Claude Code provides two web tools: WebFetchTool for retrieving content from a specific URL, and WebSearchTool for running a web search query. Both tools require permission — by default, Claude will ask you to approve each domain or search the first time it is used.

​
WebFetchTool

Fetches content from a URL and returns it to Claude, converting the response to the requested format (markdown by default).

​
Parameters

​
url
string
required
The URL to fetch. Must be a valid URL. HTTP URLs are automatically upgraded to HTTPS.
​
prompt
string
required
A description of what Claude should extract or summarize from the fetched content. The tool applies a lightweight extraction step to focus the result on what you asked for rather than returning the raw page.

​
How it works

  1. Claude sends an HTTPS request to the URL.
  2. The response body is converted to markdown (stripping HTML tags and formatting).
  3. If prompt is provided, a lightweight summarization step focuses the content on the relevant parts.
  4. The result is returned to Claude as a tool result, capped at 100,000 characters.

​
Redirect handling

If the URL redirects to a different host, WebFetchTool returns a redirect notice instead of following it automatically. The notice includes the redirect URL so Claude can make a second call with the correct address.

​
Supported content types

Content typeHandling
HTML pagesConverted to markdown; scripts and styles removed
Markdown filesReturned as-is for pre-approved hosts
PDFs and binary filesSaved to a temporary file on disk; path included in the result
WebFetchTool cannot access authenticated or private URLs (Google Docs, Confluence, Jira, private GitHub repos, etc.). For authenticated sources, use a specialized MCP server that provides authenticated access instead.

​
Permission model

The first time Claude tries to fetch from a domain, it asks for your approval. You can pre-approve domains in your local settings: 
{
  "permissions": {
    "allow": [
      "WebFetchTool(domain:docs.anthropic.com)",
      "WebFetchTool(domain:github.com)"
    ]
  }
}
Some domains are pre-approved (Anthropic documentation, common package registries). Claude does not need to ask for these.

​
Example

User: What parameters does the Anthropic Messages API accept?

Claude: I'll fetch the API reference.
[WebFetchTool: {
  "url": "https://docs.anthropic.com/en/api/messages",
  "prompt": "List all parameters for the Messages API endpoint with their types and descriptions."
}]

​
WebSearchTool

Searches the web using Anthropic’s built-in search capability and returns results with titles, URLs, and a synthesized summary.
WebSearchTool is available on Anthropic’s first-party API (claude.ai) and on Vertex AI with Claude 4.0+ models. It is not available on Amazon Bedrock or with older model versions.

​
Parameters

​
query
string
required
The search query. Must be at least 2 characters.
​
allowed_domains
string[]
When provided, only return results from these domains. Cannot be used together with blocked_domains.
​
blocked_domains
string[]
When provided, exclude results from these domains. Cannot be used together with allowed_domains.

​
How it works

WebSearchTool delegates to Anthropic’s server-side web search capability, which runs up to 8 searches internally. The results — titles, URLs, and synthesized commentary — are returned as a structured result. Claude is required to include source links in its response when using web search results.

​
Use cases

  • Looking up documentation for a library or API.
  • Finding the current version of a package.
  • Researching error messages or known issues.
  • Checking whether a feature is available in a specific version.

​
Example

User: What's the latest stable version of Bun?

Claude: Let me search for that.
[WebSearchTool: { "query": "Bun latest stable release version 2025" }]
→ Web search results for query: "Bun latest stable release version 2025"
  Bun v1.2.x is the current stable release...
  Links: [{ "title": "Bun releases", "url": "https://bun.sh/blog" }, ...]

​
Domain filtering example

// Only search the Node.js documentation
{
  "query": "fs.readFile options encoding",
  "allowed_domains": ["nodejs.org"]
}

// Exclude a noisy domain
{
  "query": "TypeScript strict mode errors",
  "blocked_domains": ["stackoverflow.com"]
}

​
Permission model

WebSearchTool requires explicit permission the first time it is used. You can pre-approve it in your local settings: 
{
  "permissions": {
    "allow": ["WebSearchTool"]
  }
}

Build docs developers (and LLMs) love

Get started for freeTalk to us