brightdata-mcp
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructions facilitate the execution of the official vendor package
@brightdata/mcpusing thenpxcommand for local deployment. - [COMMAND_EXECUTION]: Documentation includes instructions for executing shell commands such as
npx,bdata, andnodeto manage the MCP server and verify authentication status. - [CREDENTIALS_UNSAFE]: The skill describes the locations of API keys within the vendor's CLI configuration files (
credentials.json) and environment variables. However, it explicitly instructs the agent to handle these securely by never printing, echoing, or requesting the keys in cleartext within the chat. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function is fetching external web content via scraping and search tools, which introduces an inherent surface for indirect prompt injection from untrusted data.
- Ingestion points: External data is ingested via tools like
scrape_as_markdown,scrape_as_html,scrape_batch, andsearch_engine. - Boundary markers: The skill documentation recommends
scrape_as_markdownto provide the model with cleaned content, reducing the impact of malicious markup. - Capability inventory: The skill possesses network access for scraping, full browser automation capabilities (
scraping_browser_*), and search engine interaction. - Sanitization: Content is processed through the vendor's scraping engine to return parsed markdown or HTML, though specific sanitization logic is not detailed within the provided documentation.
Audit Metadata