Skip to main content
1

Install Monocle

Install via Homebrew (recommended) or see other installation methods.
brew install josephschmitt/tap/monocle
Verify the installation:
monocle --version
2

Register with your agent

Run monocle register to install skills into your agent’s config directory. With no argument, an interactive picker lets you choose which agents to register.
monocle register          # interactive picker
monocle register claude   # or: opencode, codex, gemini, all
This writes skill files into the agent-specific directory and, for Claude Code, also installs the MCP channel config for push notifications:
AgentSkills directoryMCP config
Claude Code.claude/skills/.mcp.json
OpenCode.opencode/skills/
Codex CLI.codex/skills/
Gemini CLI.gemini/skills/
Use --global to write to the user-level config directory instead of the current project.
monocle register claude --global
3

Start a review session

Run your agent in one terminal, then start Monocle in another:
monocle
Monocle connects to your agent via a Unix socket, automatically pairing based on working directory. Diffs appear in the TUI as the agent writes code.
Claude Code with push notifications: Start Claude Code with the channels flag so feedback is pushed directly into the agent’s context the moment you submit:
claude --dangerously-load-development-channels plugin:monocle@monocle
The --dangerously-load-development-channels flag is required during the channels research preview. If you restart Monocle while Claude Code is already running, type /mcp in Claude Code and select Monocle to reconnect.
4

Leave comments and submit

Navigate diffs with j/k and add a comment with c. Use v to select a range of lines before commenting. Choose a comment type (Tab cycles through issue, suggestion, note, praise) and press Enter to save.When you’re done reviewing, press S to submit. Your review is formatted and queued for delivery:
  • Claude Code with channels: a push notification delivers feedback immediately
  • All other agents: feedback waits in the queue until the agent runs /get-feedback
If you submit with no comments, the review is treated as an approval.
Press ? at any time to see all available keybindings, or see the full keybindings reference.

Next steps

Review loop

Learn the full review cycle: pause flow, feedback queue, file tracking, and submission history

Plan review

Have your agent submit plans and architecture decisions for review before writing code

Agent integration

Configure Claude Code push notifications, socket pairing, and session modes

Keybindings

Full reference for all TUI and comment editor keybindings

Build docs developers (and LLMs) love