Executive brief
Claude Code is an AI-powered development assistant that performs web requests to fetch documentation and resources. A flaw in how the application validates trusted domains allows attackers to register lookalike domains (e.g., modelcontextprotocol.io.attacker.com) that pass the validation checks. This could trick Claude Code into automatically fetching content from attacker-controlled servers, potentially exposing sensitive data without user awareness.
Technical details
Claude Code's WebFetch mechanism uses a flawed URL validation check—a simple startsWith() function—to verify that requests go only to trusted domains like docs.python.org and modelcontextprotocol.io. An attacker can bypass this by registering a domain that starts with a trusted domain name (e.g., modelcontextprotocol.io.attacker.com), which would pass the validation. No authentication is required; the vulnerability is triggered when Claude Code automatically makes a request to the attacker-controlled domain, potentially resulting in data exfiltration. The attack vector is network-based and requires only passive user interaction (e.g., viewing a prompt or file that triggers the WebFetch). Patches are available as of version 1.0.111, and users on auto-update have already received the fix.
Affected products
- Anthropic Claude Code < 1.0.111
Timeline
- 2026-02-03: disclosed
- 2026-02-03: patched: Fix released in version 1.0.111