Claude Code WebFetch 403: why it fails and how to fix it

Three different things can stop WebFetch. Only one of them is the site. Check this first, because each cause has a different fix.

1. 1

   Read the error text

   A block by the site shows an HTTP status: 403, 401, 402, or 429. A block by Claude Code does not. It shows a permission prompt that was denied, or a message that Claude Code could not verify the domain.

2. 2

   Ask curl for the same page

   Run curl on the same machine. If curl also gets 401 or 403, the site refuses requests that do not come from a browser.

   Step 2

   curl -sL -o /dev/null -w "%{http_code}\n" https://www.reuters.com/technology/

3. 3

   Open the page in your browser

   If the browser shows the page and you are not logged in, the page is public. Only the request of the tool is refused. This is the case that this guide fixes.

WebFetch sends its request from your computer. A site with bot protection looks at that request before it answers. It checks the name the client gives, the TLS handshake, the HTTP/2 settings, and the order of the headers. A browser has one known pattern for each. WebFetch gives the name Claude-User, and the rest of its request is the pattern of a program, not of Chrome or Safari.

Your IP address is not the main cause. In our test, WebFetch, curl, and Anyfetch ran on the same server with the same address. WebFetch read 36 sites, and Anyfetch read 82.

The status code tells you what the site wants. WebFetch got 403 from 33 of the sites, 402 from 6, and 401 from 2. A 402 means “payment required”: the site asks AI crawlers to pay for access. Some sites send 200 with nothing useful. From Reddit, WebFetch got a page with only the word Reddit. From Amazon, it got the menu and no products.

Anyfetch is a free, open-source MCP server from the AliasMode team. It gives Claude Code three tools: web_search, web_fetch, and web_fetch_many. Its requests are the requests that a browser sends, so most sites answer with the page. No browser starts, and you need no API key.

curl -fsSL https://aliasmode.github.io/anyfetch/install.sh | sh

irm https://aliasmode.github.io/anyfetch/install.ps1 | iex

1. 1

   Run the installer

   It puts one program in ~/.local/bin and adds it to Claude Code as an MCP server. It does the same for Codex, Gemini CLI, Cursor, and the other agents it finds.

2. 2

   Or install it as a plugin

   On macOS and Linux you can use the Claude Code plugin in place of the installer. Use one of the two, not both.

   Step 2

   claude plugin marketplace add aliasmode/anyfetch && claude plugin install anyfetch@anyfetch

3. 3

   Or add it by hand

   With Node.js on the machine, this line needs no install at all.

   Step 3

   claude mcp add --scope user anyfetch -- npx -y @aliasmode/anyfetch mcp

4. 4

   Start a new session and test

   Run claude mcp list and look for anyfetch. Then ask for a page that failed before, for example: “Read https://www.reuters.com/technology/ and list the first three headlines.” Claude calls web_fetch and gets the page.

Claude still has WebFetch and can choose it. To make Anyfetch the only web tool, deny the two built-in tools in ~/.claude/settings.json:

{
  "permissions": { "deny": ["WebFetch", "WebSearch"] }
}

Anyfetch reads what a site sends to a first-time visitor. It does not solve a CAPTCHA, log in, or open a page behind a subscription. 15 of the 97 sites stayed closed in our test. Booking.com, TikTok, Crunchbase, and G2 are among them. For those, Anyfetch returns an error code such as challenge_required, so Claude knows that it must not guess.

Does Claude Code run on a server? Some sites refuse every cloud address, whatever the request looks like. A residential IP fixes those. With one NobleProxy address, Anyfetch read 87 of the 97 sites. The Wall Street Journal, Tripadvisor, Yelp, Etsy, and Bandcamp opened.

Does the task need a login, a click, or a form? Then the agent needs a real browser. AliasMode gives Claude Code browser profiles with saved sessions through its own MCP server.