crimeapp-mcp

CRITICAL: The `isAuthorized` checks for `/proxy/db/query` and `/proxy/news_articles` endpoints in `crimeapp-mcp-server/src/index.ts` are commented out, making these internal database proxy endpoints publicly accessible without authentication. While `sanitizeSelect` limits queries to `SELECT` and adds `LIMIT 1000`, this still allows for unauthenticated information leakage of sensitive crime data. Additionally, the `AGENT_SHARED_SECRET` check in `crimeapp-mcp-server/src/lib/agent.ts` is commented out, meaning calls to the external LangChain agent (`LANG_AGENT_URL`) are unauthenticated, which is another major security flaw. The `lang-agent` generates SQL queries using an LLM (`_plan_sql_from_query`), which, despite SQL sanitization (`_sanitize_select`), introduces a high risk of LLM hallucinating or being prompted to generate malicious or resource-intensive queries that could bypass imperfect guardrails.