Claude/Code/Q/What can I do in Claude Code (CLI) that I can't do in the VS Code extension?

• What can I do in Claude/Code (CLI) that I can't do in the VS Code extension?

  • AI Answer

    • Short answer: The VS Code extension is built for interactive, in-editor coding and intentionally omits the terminal-native, scripting, and automation surface of the CLI. The extension also bundles its own copy of the CLI for its chat panel; to get the full CLI inside VS Code you run the standalone install in the integrated terminal. The genuinely CLI-only or CLI-advantaged capabilities cluster into the groups below.
    • Headless and scripting
      • Print/query mode (claude -p "…") that runs once and exits — the extension is interactive chat only.
      • Piping: cat file | claude -p "query" to feed content without manual input.
      • Machine-readable output via --output-format json / stream-json, with --json-schema for validated structured output.
      • --bare for fast scripted startup (skips hooks, skills, plugins, MCP discovery).
    • Background and parallel sessions
      • Background agents (--bg) that return a session ID for CI/CD, managed with claude attach / logs / stop / respawn / rm.
      • Git worktree isolation with terminal multiplexers: --worktree plus --tmux for parallel environments.
      • One-shot background commands: --bg --exec 'pytest -x'.
    • Session control and resumption
      • claude --resume picker with search, -c to continue the latest session in the current directory.
      • --fork-session to branch mid-conversation, and --from-pr <number> to resume all sessions tied to a PR.
      • --session-id <uuid> for deterministic session tracking.
    • Terminal-native UX
      • Tab completion, the ! bash shortcut for shell commands, and Shift+Tab permission-mode cycling.
      • The full slash-command surface — the extension exposes only a subset.
    • MCP and configuration per invocation
      • Shell management: claude mcp add / login / logout, headless OAuth via claude mcp login <server> --no-browser.
      • Per-run config: --mcp-config, --strict-mcp-config, --settings ./settings.json, --setting-sources, --add-dir.
    • Automation and cost control
      • --max-turns, --max-budget-usd hard caps for scripted runs.
      • Per-invocation --permission-mode, --allowedTools, --disallowedTools, --model / --fallback-model, --effort.
      • System-prompt control: --system-prompt(-file), --append-system-prompt(-file).
    • Remote and web bridge
      • --remote creates a new claude.ai web session from the terminal, --teleport pulls a web session into the local terminal, --remote-control lets the Claude app drive a local run.
    • Auth, plugins, and debugging
      • CI auth: claude auth login, claude setup-token for long-lived tokens; Bedrock/Vertex/Foundry routing via env vars.
      • Plugin lifecycle from the shell: claude plugin install / list / remove, plus per-run --plugin-dir / --plugin-url.
      • --safe-mode, --debug [categories], --debug-file, claude project purge.
    • Caveat: Claude Code changes fast — some of these may land in the extension over time. The docs maintain a current comparison: VS Code extension vs. CLI, full CLI reference, and feature availability by platform.