cdp-bridge-mcp-browser-control

Warn

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION]: Provides the browser_cookies tool, which allows the agent to read session tokens and sensitive data from the user's active, authenticated browser sessions.
  • [COMMAND_EXECUTION]: Includes the browser_execute_js tool, enabling the execution of arbitrary JavaScript code within the context of any open browser tab.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The browser_scan tool reads untrusted content from the current webpage (DOM) directly into the agent's context.
  • Boundary markers: No explicit delimiters or instructions are provided to the model to ignore instructions embedded within the scanned webpage content.
  • Capability inventory: The agent has access to powerful capabilities (cookie access, JS execution, navigation) that can be misused if the model is compromised by content it reads.
  • Sanitization: The skill performs HTML filtering for token reduction but does not mention sanitizing content to prevent prompt injection.
  • [EXTERNAL_DOWNLOADS]: The installation process requires downloading and executing the cdp-bridge package from PyPI via uvx and loading an unpacked extension from a local source.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 16, 2026, 10:42 PM
Security Audit — agent-trust-hub — cdp-bridge-mcp-browser-control