Phoenix — AI skill for Claude Code

Phoenix Memory

Persistent memory for AI coding CLIs with **zero token overhead**.

Works with **Claude Code**, **Codex**, and **Gemini CLI** — all three support both MCP tools and auto-logging hooks.

---

How it works

Claude Code / Codex / Gemini CLI
  │
  ├── MCP tools  (model calls these on purpose)
  │     initialize_session        — start or warm-start a session
  │     create_manual_checkpoint  — save a milestone summary
  │     get_resume_packet         — reload context from a previous session
  │     get_handoff_packet        — export conversation for another CLI
  │
  └── Hooks  (runtime fires these automatically, 0 tokens)
        PostToolUse / AfterTool  →  phoenix-log  — saves every tool call + result
        Stop / SessionEnd        →  phoenix-log  — saves transcript snapshot
              │
              ▼
        SQLite  (~/.phoenix_memory/state.db)

---

Install

Requires **Python 3.11+** ([download](https://www.python.org/downloads/))

git clone git@github.com:khangbdd/phoenix.git && cd phoenix
./setup.sh
source ~/.zshrc   # or restart your terminal

After that `phoenix-server`, `phoenix-log`, and `phoenix` work globally.

---

Wire up your CLI

Claude Code

**MCP server** — `~/.claude.json`:

{
  "mcpServers": {
    "phoenix-memory": { "command": "phoenix-server" }
  }
}

**Hooks** — `~/.claude/settings.json`:

{
  "hooks": {
    "PostToolUse": [
      { "matcher": ".*", "hooks": [{"type": "command", "command": "phoenix-log post_tool_use"}] }
    ],
    "Stop": [
      { "hooks": [{"type": "command", "command": "phoenix-log stop"}] }
    ]
  }
}

Codex

**MCP server** — `~/.codex/config.toml`:

[mcp_servers.phoenix-memory]
command = "phoenix-server"
args = []

**Enable hooks feature**:

codex features list | rg codex_hooks

If the command above returns nothing/false, run:

codex features enable codex_hooks

**Hooks** — `~/.codex/hooks.json`:

{