codeparse MCPサーバー

Indexing Python codebases into a local SQLite database allows AI agents to access compact context while minimizing token consumption. This server parses your code to map relationships and stores them locally, making it an efficient way for developers to provide large repositories to LLMs without overwhelming the context window.

Users set up the server by placing the codeparse binary on their PATH and running the install command. Once active, the server provides specific tools like get_symbol, search_symbols, and get_import_relations to help agents navigate complex structures. It also includes a guided exploration skill and a resource to track tokens saved during a session.

Developers can manage the index using commands such as reload to reprocess a workspace or update to refresh the binary. By using tools like get_file_symbol_tree and list_subclasses, an agent can understand class hierarchies and module dependencies through the structured SQLite backend rather than reading every line of raw text.