cloudflare-mcp

Pass

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses standard platform tools such as wrangler and pnpm for deploying Cloudflare Workers. It also includes a maintenance script (scripts/delta.py) that utilizes the GitHub CLI (gh) to query official repositories for platform updates.
  • [EXTERNAL_DOWNLOADS]: The skill's maintenance workflow fetches data from official and well-known registries including registry.npmjs.org and official GitHub organizations like cloudflare and modelcontextprotocol. These operations are used solely to synchronize the skill's technical guidance with upstream developments.
  • [INDIRECT_PROMPT_INJECTION]: The maintenance process (described in references/refresh.md) involves triaging data from external sources like RSS feeds and GitHub commit messages. This potential attack surface is addressed by strict instructions for human-in-the-loop verification and triage, ensuring that no untrusted content is automatically incorporated into the skill's core logic.
  • [DYNAMIC_EXECUTION]: The skill's documentation describes how to implement 'Code Mode' servers using the Cloudflare @cloudflare/codemode library. This involves executing sandboxed code via the platform's DynamicWorkerExecutor, which is a legitimate and documented feature of the target development environment.
  • [CREDENTIALS_UNSAFE]: Security is prioritized in configuration examples; sensitive values like API keys and encryption secrets are handled using placeholders and environment variable instructions (e.g., wrangler secret put), avoiding hardcoded credentials.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 16, 2026, 05:21 PM