Clancey
Clancey is a shared conversation index for AI coding tools. It imports what you already did in Claude Code, Grok Build, OpenCode, Codex, and Hermes into one place, so any of those agents can answer:
- "Which conversation produced the
feature/authbranch?" - "What did we say the last time we touched
GameRepository.ts?" - "Why did we move auth to the edge — and in which client was that decided?"
You ask your current agent; it looks up the session across every client Clancey knows about, then reads back what was actually said.
Supported tools
One store (~/.clancey/). Setup wires each tool’s MCP server and installs the Clancey skill so agents know how to look things up. Live capture of file edits and shell commands is silent infrastructure for branch/file mapping — it does not coach the agent.
Optional: as you work, the skill can also have the agent record decisions and learnings so semantic search is richer later. Lookup works without that; imported transcripts are enough.
Prerequisites
- Node 18+
- At least one of the supported coding tools above
Install and set up
npx -y clancey setup
Setup asks which tools to enable, imports existing history from each so you can ask about it immediately, and installs the Clancey skill. Restart your tools when it finishes. The first run downloads a small embedding model (~30 MB), cached afterward.
Using it
You never call Clancey directly. Your agent does, whenever you ask about past work — including work that happened in a different coding client than the one you’re in now.
Things you can ask:
- "Which conversation produced this PR?" / "…the
feature/authbranch?" - "What was I thinking the last time I changed
GameRepository.ts?" - "What did we do in this repo recently?"
- "What did Codex do on
featlast week?" (or Claude, Grok, OpenCode, Hermes) - "Search everything we said about edge auth a week ago." / "…yesterday." / "…in the last 7 days."
Clancey can scope by repo (checkout path or owner/name), branch, which coding tool, and plain-English time (“last week”, “a week ago”, “yesterday”). It can list recent sessions, open the full transcript (including subagent turns), and fall back to a snapshot if the original chat file was pruned.
While any coding client has Clancey MCP open, a single background clancey watch keeps the index fresh by re-ingesting changed transcripts. You do not need to run backfill by hand for ordinary work. If something still looks missing, ask the agent to refresh the index (or run npx clancey backfill).
Optional enrichment: as you work, the agent may record decisions and learnings (commits, PRs, root causes, approach choices) so later search is sharper. You can also ask it to mine older sessions for decisions after the fact. Lookup works without that — imported history is enough.
Keep your history
[!IMPORTANT] Claude Code deletes chat transcripts after 30 days by default, so do this early. Clancey snapshots the conversations it imports, so anything it has already seen survives pruning, but raising the window keeps conversations around long enough to be imported in the first place:
{ "cleanupPeriodDays": 3650 }
Commands
You only ever run setup by hand; your agent runs everything else for you. Setup pins the MCP server, skill, and live capture to this install, so a later release won't change your setup until you re-run it.
npx clancey setup Set up Clancey and import history (run once)
npx clancey backfill Re-import / refresh the conversation index
npx clancey watch Background indexer (auto-started by MCP; one instance)
Your agent can also refresh the index from inside a session when results look stale. Run npx clancey --help for the full list. Everything Clancey stores lives in ~/.clancey/.
Development
bun install
bun run typecheck
bun run test
bun run build
License
MIT
