mcp-files is an MCP server that enables AI agents to rapidly locate and surgically edit code across a codebase. It exposes a suite of type-safe tools for symbol-level code operations, including read_symbol for cross-repository symbol discovery, insert_text for precise edits, import_symbol for module inspection, search_replace for resilient text modifications, and os_notification for system alerts. It supports multiple transport modes (stdio, HTTP) and can be run via NPX, Docker, or an HTTP transport, with environment variables to configure transport, port, and debug mode. The server emphasizes cross-codebase intelligence—finding symbols anywhere with exact locations—and surgical editing that targets exact functions, classes, or blocks without disturbing surrounding code. It supports zero-error refactoring by enabling changes to API surfaces across files. The architecture uses type-safe tools with Zod validation, self-contained modules, and cross-platform support, with memoization and descriptive error messages for performance and reliability. A dedicated code-optimization feature can strip comments, normalize indentation, collapse blank lines, and remove common leading indentation to shrink AI-context token usage. CLI examples and deployment options cover NPX, Docker, and HTTP transport.