Ccpeek

CCPeek

Explore your coding-agent history. A local web app that indexes
Claude Code, Pi, Codex CLI, and OpenCode sessions (plus
Cursor, experimental — see the capability matrix below) into one
session-centric database — conversations, plans, todos, tasks, shell
snapshots, file history, paste cache, memories, and commands — with real
token usage and estimated cost, plus an agent-facing query surface
(ccpeek query, /api/v1, ccpeek mcp) and live updates. Everything
stays on this machine.

Installation

Homebrew

mise

Nix Flakes

Build from source (full product)

Requires Go 1.25+, Node.js, pnpm, curl, jq, and Playwright's Chromium browser:

just build refreshes the embedded pricing snapshot, runs the full test suite against it, and then compiles the product with the withui build tag. The tag enforces the embedded UI's presence at compile time, so just build (and every release path) cannot produce a UI-less binary.

go install (API-only variant)

A plain go build ./... or go install cannot run the SPA build, so
it deliberately produces the API-only variant: /api/v1,
ccpeek query, and ccpeek mcp work normally, the server logs a
warning at startup, and / explains what is missing instead of
rendering a blank page. Use it for headless/agent-only setups; use any
other installation method for the web UI.

Pre-built binaries

Download from GitHub Releases.
Archives include shell completions and man pages.

Usage

The server reads each agent's data from its default root (for Claude Code,
~/.claude), writes an index to $XDG_DATA_HOME/ccpeek/ccpeek2.db, and
serves the web UI at http://localhost:3000.

The port binds immediately; indexing runs behind it with progress on
stderr, and the UI fills in live as data lands (/api/v1/ready answers
200 once the first pass completes). After…