paulbenchea/codelore

CodeLore MCP is a server that gives an AI coding assistant a structured place to store and retrieve documentation about a codebase, so the assistant can read a short summary and fetch exactly the doc it needs instead of re-reading the whole codebase every time. Every piece of documentation comes in two versions: an internal doc that explains how the code works, and a usage doc that explains how to use it, with inputs, outputs, and examples. Documentation is organized in a folder hierarchy stored on disk, by default under a hidden folder in the user's home directory. Within a project, documentation is grouped by category, such as a technology like Angular or SQL, then by chapter, which is a functional area within that category, and finally by topic, which is the actual unit of documentation holding the internal and usage files. The server enforces this structure itself: names are automatically formatted consistently, each document gets standard metadata and a generated title, and the index files that list what exists are regenerated from disk on every write so they cannot fall out of sync. The AI assistant only ever needs to supply the actual written content. The server exposes a set of tools an assistant can call: listing and registering projects, defining categories, chapters, and topics, writing or reading a topic's internal or usage documentation, searching across all documentation with fuzzy and prefix matching, and getting a full map of a project's table of contents in one call. The intended flow is to build up documentation while working with a user, then later read back just the table of contents or a specific topic when doing a new task, or search by keyword when the right topic is not known in advance. Once published to npm, it can be registered with Claude Code using a single command, or added to any MCP client's configuration file with npx. Where documentation is stored can be customized with a root flag or an environment variable instead of using the default home directory location.