firecrawl-mcp

This module is not itself demonstrably malicious: it lacks dynamic code execution, network code, or obfuscation in this file. The primary security risks are operational: a hardcoded default access key, lack of payload sanitization, and forwarding of arbitrary JSON or file contents to an external, un-audited router implementation (ai_agent_marketplace.OneKeyAgentRouter). Treat payloads as highly sensitive; avoid running with the default baked-in key; and audit the OneKeyAgentRouter implementation to determine whether the invoke call exfiltrates data or executes untrusted code. If the router is untrusted, do not pass secrets or private files to this CLI.

The firecrawl-mcp skill presents a coherent mapping between its stated purpose (unified access to web-scraping/crawling tools via OneKey Router) and its capabilities (multiple scraping/crawl/agent/browser tools). The primary concerns are credential handling (usage of an API key with a demo fallback and lack of rotation/scoping), network/data flow exposure (external API calls and large potential data transfers), and autonomous action risk (async agent workflows requiring governance). Overall, the footprint is proportionate to a sophisticated web-research/tooling suite but carries moderate security risk due to credential exposure potential and broad data/network flows. Recommended mitigations include explicit credential scoping, per-tool permissions, explicit TLS verification expectations, rate limits, audit logging, and user confirmations for high-risk autonomous actions.