I've been using Claude Code daily for ~1 year now. It's my default LLM interface, not just for coding, but for all types of work. In a given month, I'd have hundreds of sessions across different projects and topics. Frequently, I'd want to inject context from a past session into my current one, but it was annoying to find that old session, locate the relevant part, and then copy + paste it.

/resume (Claude Code's session picker) lets you search past sessions, but you can only filter by first message or custom name. Otherwise, you're scrolling through a time-sorted list. This is painful when you have a ton of sessions. You can grep the JSONL files (conversation logs) if you remember the exact words, but I often didn't, or the exact words would yield too many session results.

claude-find is an MCP server that indexes your Claude Code sessions locally. It chunks the conversations, enriches them with metadata, embeds them on your GPU with Ollama, and stores everything in SQLite. It combines semantic matching with keyword matching for search. It returns the raw conversation with the full reasoning, not just a summary. You can type /find in any session and ask a question in plain language. Claude Code deletes sessions after 30 days by default, but the setup disables this cleanup, so nothing expires.

It works on all existing sessions. Embeddings run locally, so nothing leaves your machine.

I like it, I'll check it out today