code-memory

The server's core functionality involves indexing and searching local file systems (`index_codebase`, `search_code`) based on user-provided paths. While `path.resolve` normalizes paths, there are no explicit sandboxing mechanisms to restrict the 'path' argument to a specific, safe directory (beyond the `.code-memory` data storage itself). This means an attacker who can manipulate the AI's prompts could potentially coerce the MCP server to read arbitrary files from the local file system (e.g., sensitive configurations, private keys, etc.) that the Node.js process has access to. This is a significant arbitrary file read vulnerability if the AI client is compromised or jailbroken. No `eval` or intentional obfuscation was found, and error handling is present, but the powerful file access is a critical risk.