# Claude Code CLI Guide

Claude Code is Anthropic's agentic coding tool. It reads your codebase, edits files, runs commands, and integrates with dev tools. Available as a terminal CLI, VS Code/Cursor extension, desktop app, and in-browser.

**Official docs:** https://code.claude.com/docs/en/overview
**CLI reference:** https://code.claude.com/docs/en/cli-reference

---

## Installation

```bash
# macOS / Linux / WSL (recommended — auto-updates)
curl -fsSL https://claude.ai/install.sh | bash

# Homebrew (does NOT auto-update)
brew install --cask claude-code

# npm (global)
npm install -g @anthropic-ai/claude-code
```

**Installed on:** Tabitha, Hunter's machine, Ozymandias

---

## Authentication

```bash
claude auth login              # Sign in with Claude subscription (claude.ai)
claude auth login --console    # Sign in with Anthropic Console (API billing)
claude auth status             # Check current auth state (JSON)
claude auth status --text      # Human-readable
claude auth logout
```

---

## Core CLI Commands

```bash
claude                              # Start interactive session
claude "query"                      # Start session with initial prompt
claude -p "query"                   # One-shot query, then exit (SDK mode)
cat file.ts | claude -p "explain"   # Pipe content in
claude -c                           # Continue most recent conversation
claude -c -p "query"                # Continue + one-shot
claude -r "session-name" "query"    # Resume session by ID or name
claude update                       # Update to latest version
```

---

## Key Flags

| Flag | Description |
|------|-------------|
| `--model <model>` | Specify model (e.g. `claude-opus-4-5`, `claude-sonnet-4-6`) |
| `--add-dir <path>` | Add directory to allowed paths |
| `--permission-mode` | Set permission mode (`auto`, `default`, `manual`) |
| `--output-format` | `text` (default), `json`, `stream-json` |
| `--max-turns <n>` | Limit agentic loop turns |
| `--no-update-notifier` | Suppress update messages |
| `--mcp-config <file>` | Load MCP config from file |
| `--settings <file>` | Load settings from JSON file |

---

## In-Session Slash Commands

```
/help           Show available commands
/status         Show auth, model, context info
/model          Switch model mid-session
/clear          Clear conversation history
/compact        Summarize conversation to save context
/memory         Show loaded memory files (CLAUDE.md etc.)
/tools          List available tools
/reasoning      Toggle extended thinking on/off
/cost           Show token usage for session
/quit           Exit
```

---

## MCP Servers (Tabitha — User Scope)

Configured in `~/.claude.json`. All available in every Claude Code session on Tabitha:

```bash
claude mcp list                                          # Show all MCPs
claude mcp add --scope user -e KEY=val -- name npx -y pkg  # Add one
claude mcp remove --scope user name                      # Remove one
```

| MCP | What it gives Claude |
|-----|---------------------|
| `github` | GitHub API (silmaai PAT) |
| `postgres` | SQL → silma DB |
| `memory` | Knowledge graph (`memory/knowledge-graph.json`) |
| `redis` | Redis client |
| `qdrant` | Vector search on `silma` collection |
| `elasticsearch` | Full-text search |
| `minio` | S3-compatible object storage |
| `playwright` | Headless browser automation |
| `kubernetes` | `kubectl` via MCP |
| `google-workspace` | Gmail, Drive, Docs, Sheets, Calendar |

> ⚠️ postgres, redis, qdrant, elasticsearch, minio require k8s port-forward active.

---

## CLAUDE.md (Project Memory)

Claude Code automatically loads `CLAUDE.md` from the project root and parent directories. This is how you give Claude project-specific instructions that persist across sessions.

Create one at the project root:
```bash
echo "# Project Instructions\nAlways use TypeScript strict mode." > CLAUDE.md
```

On Tabitha, Silma's workspace CLAUDE.md equivalent is `AGENTS.md`.

---

## Using Local LLMs with Claude Code

Claude Code supports OpenAI-compatible third-party providers. Point it at the local llama.cpp server:

```bash
# Set provider to local llama.cpp (Gemma models on Tabitha)
export ANTHROPIC_BASE_URL=http://localhost:8080/v1
export ANTHROPIC_API_KEY=local   # required but ignored by llama.cpp

claude --model gemma-4-e4b
```

See `tools/local-llm.llm.md` for full local LLM details.

---

## Environment in This Workspace

| Detail | Value |
|--------|-------|
| **Workspace** | `/Users/Tabitha/.openclaw/workspace/` |
| **OCPlatform channel** | webchat (main session) |
| **Model** | `anthropic/claude-sonnet-4-6` (check `/status`) |
| **Agent YAML** | `AGENTS.md` at workspace root |
| **Memory** | `memory/YYYY-MM-DD.md` (daily) + `MEMORY.md` (long-term) |

---

## Tips for This Environment

- `Glob` and `Grep` built-in tools are **not available** in OCPlatform sessions — use `Bash` with `find`/`grep` instead
- For cross-machine work, use the kittens WebSocket bridge (see `tools/tailscale.llm.md`)
- Run `source credentials/secrets.env` inside Bash tool calls to load credentials
- k8s MCPs require port-forward — run `bash k8s/silma-ai/scripts/port-forward.sh` first