mcp

[Skill Scanner] Agent capability enumeration detected This skill/documentation appears to be benign instructional material for building MCP servers and clients. It does not contain obfuscated or purposely malicious code. However, several example patterns are unsafe if copied verbatim: unsanitized SQL (insert_record), arbitrary file writes/reads (write_file/read_file), and unvalidated URL fetching (fetch_url). There is also a minor supply-chain consideration from npx-based tooling. Recommend marking as medium security risk for potential misuse and ensuring examples are hardened before use (validate/normalize paths, sanitize table names, use parametrized queries, avoid hardcoding secrets, validate URLs, and avoid writing secrets to logs). LLM verification: [LLM Escalated] This is legitimate MCP SDK documentation and examples that demonstrate powerful capabilities needed for LLM context integration (file IO, DB, HTTP, subprocess transports). I found no direct indicators of malware or obfuscated malicious code. However, several example patterns are high-risk if copied into production without proper authorization, input validation, and supply-chain hygiene: arbitrary filesystem access, interpolated SQL object names, subprocesses inheriting host environment, and unpi