claude-session-search 0.1.0

Blazing-fast local full-text search (SQLite FTS5) for your Claude Code sessions

cc·search

Blazing-fast local full-text search for your Claude Code sessions.

Every conversation you've ever had with Claude Code — your prompts, Claude's answers, its internal thinking, every tool call and its output, even subagent transcripts — indexed and searchable in milliseconds.

• Zero dependencies — pure Python standard library (SQLite FTS5 is built in)
• 100% local & private — binds to 127.0.0.1 only, makes zero network requests, nothing ever leaves your machine
• Fast — indexes ~200k messages (1.8 GB of transcripts) in under 10 seconds; incremental after that

Installation

Run it instantly with uv (nothing installed permanently):

uvx claude-session-search

Install it with pipx or pip:

pipx install claude-session-search    # or: pip install claude-session-search
cc-search

Or straight from GitHub:

uvx --from git+https://github.com/aleducode/cc-search cc-search

Or clone and run:

git clone https://github.com/aleducode/cc-search && cd cc-search
./run.sh

Any of the above indexes your sessions and opens http://localhost:7777 in your browser. The first run takes a few seconds; subsequent runs only process files that changed.

cc-search --port 8080     # use a different port
cc-search --no-open       # don't open the browser
cc-search --reindex       # rebuild the index and exit

The index lives at ~/.cc-search/index.db.

What gets indexed

Everything inside every session under ~/.claude/projects/:

Role	Content
user	Your prompts
assistant	Claude's responses
thinking	The model's internal reasoning
tool_use	Every tool invocation and its arguments
tool_result	Tool output (truncated to 4 KB per message)

Subagent transcripts (<session>/subagents/*.jsonl) are included too — tagged with ⑂ and attributed to their parent session.

Search features

• SQLite FTS5 full-text engine with BM25 relevance ranking
• Diacritic-insensitive: migracion matches migración
• Automatic prefix matching: index matches indexer, indexing, …
• Exact phrases with quotes: "connection refused"
• Filters: project (sidebar), message role (chips), date range, sort by relevance or recency
• Full-context view: click any result to open the entire conversation with the matching message centered and highlighted
• Keyboard-first: / focuses the search box, Esc closes panels

How it works

cc_search/indexer.py     JSONL parser → SQLite (~/.cc-search/index.db) with an FTS5 virtual table
cc_search/server.py      stdlib HTTP server + JSON API: /api/search, /api/sessions,
                         /api/session/<id>, /api/projects, /api/stats, /api/reindex
cc_search/static/        single-page UI, no build step

The indexer tracks each file's mtime and size, so reindexing (on startup or via the ⟳ Reindex button) only reprocesses sessions that changed.

Privacy

Everything is strictly local. The server listens on 127.0.0.1 only and performs no outbound requests of any kind. Your transcripts never leave your machine.

License

MIT