What is Plannotator?
Plannotator opens a review page in your browser when a coding agent proposes a plan, writes a document or finishes a change. You highlight text, comment, suggest code on a diff, approve or send back, and the feedback goes straight to the agent session. It hooks into plan mode for Claude Code, Codex, Copilot CLI, Gemini CLI, OpenCode, Kiro, Droid, Amp, DeepSeek Harness and Pi, and adds commands to annotate any Markdown file, folder, HTML page or URL, or the agent's last message.
Code review covers local changes and remote pull or merge requests, with Git, GitButler, Jujutsu, Perforce, GitHub and GitLab. The newer Inbox is one local window where several agents leave questions, files and reviews while they keep working; agents without a plugin reach it over MCP. Herdr Annotate brings the same review into the terminal.
The project is dual licensed MIT or Apache-2.0, started in December 2025 and ships releases most days; v0.28.64 came out on 13 November 2026. A hosted team product, Workspaces, has a waitlist.
Is Plannotator worth it?
Plan mode is only as good as your review of the plan, and a long plan scrolling past in a terminal is easy to approve unread. Plannotator puts it in a proper page where you can strike a step, comment on a line and send the whole set back in one go. The same surface handles diffs, so the loop from plan to change to review stays in one tool and the agent receives structured feedback instead of a paragraph typed from memory.
It is busy and broad: 10,865 stars, 808 forks, merges on the morning of 14 November 2026 and three releases on 12 and 13 November. One installer detects the agents you have and wires hooks, skills and slash commands for each, across ten coding agents. On Claude Code 2.1.287 or newer, reviews no longer hold the session open; Claude ends its turn and your decision arrives later as a message.
The README is more precise about network behaviour than most commercial tools. There is no telemetry, and plans, diffs and annotations stay local. Each review page does check GitHub for a new release when it loads, with no opt-out. Small shared links carry the plan unencrypted in the URL fragment; large shares are encrypted with AES-256-GCM before upload and expire after 7 days, and sharing can be switched off.
The catch: the recommended install is a curl script piped to bash that changes configuration for every agent it finds, so read it or use the minimal flag. Release pace is fast enough that plugin and binary versions can drift; the README spells out separate update steps per agent. Link sharing is being deprecated in favour of the hosted Workspaces, which has no price yet.
Our take: anyone who approves agent plans more than a few times a day should have this installed. It turns review from a chore you skip into the step where your judgement actually lands.
Where it earns its place
A senior engineer delegating features to Claude Code
Review a long implementation plan before the agent starts editing
With the plugin installed, plan mode opens the plan in the browser; strike the steps you disagree with, comment on the rest and send it back. Free; nothing leaves your machine.
A tech lead reviewing a teammate's agent-written pull request
Leave line comments on a GitHub pull request and hand them to an agent to fix
Run /plannotator-review with the pull request URL; it reads the diff through your authenticated GitHub CLI, and your comments go to the agent session as its next task.
A developer running four agent sessions at once
Answer every agent's questions in one place without blocking any of them
Start plannotator inbox; Claude Code, Pi and OpenCode connect through plugins and other agents over MCP, and each reply returns to the session that asked.
Before you commit
- Setup
- Run the install script for macOS, Linux or WSL (or the PowerShell one on Windows), then the agent-specific step, such as adding the plugin marketplace in Claude Code; --minimal installs only the binary.
- Integrations
- Claude Code, Codex, Copilot CLI, Gemini CLI, OpenCode, Kiro, Droid, Amp, DeepSeek Harness and Pi; T3 Code; MCP for other agents; GitHub and GitLab through their CLIs; Herdr in the terminal.
- Data and privacy
- No telemetry. Content stays local except URL annotation (through Jina Reader by default), pull request fetches, AI questions sent to your chosen provider, and shares; PLANNOTATOR_SHARE=disabled turns sharing off.
- Limits
- Some agents get commands only, without plan interception; Codex hooks on native Windows are experimental; hosted pastes expire after 7 days.
What’s good
- Annotate plans, documents, HTML and diffs, then send feedback as the agent's next turn
- Works with ten coding agents, wired by one installer
- No telemetry; content stays local unless you share or ask a model
- Large shares encrypted with AES-256-GCM and expire after 7 days
- Code review across Git, GitButler, Jujutsu, Perforce, GitHub and GitLab
What’s not
- Default install pipes a script to bash and edits several agents' configs
- Release check to GitHub on every review page load, with no opt-out
- Small share links carry content unencrypted in the URL
- Fast release pace means per-agent update steps to keep in sync
Best for
Developers who use plan mode in Claude Code, Codex or similar agents every day and want to review plans and diffs with the care they give a pull request.
Plannotator alternatives
Plannotator: common questions
- Is Plannotator worth it?
- Plannotator fixes the weakest moment in agent coding, reading a long plan in a terminal, and does it locally with unusually candid privacy notes. We rate it 4.2 out of 5.
- How much does Plannotator cost?
- Free and open source under MIT or Apache-2.0. Workspaces, a hosted team layer, is on a waitlist with no published price. AI questions inside reviews use your own agent or model provider.
- What are the downsides of Plannotator?
- Default install pipes a script to bash and edits several agents' configs; release check to GitHub on every review page load, with no opt-out; small share links carry content unencrypted in the URL; fast release pace means per-agent update steps to keep in sync.
- Who is Plannotator best for?
- Developers who use plan mode in Claude Code, Codex or similar agents every day and want to review plans and diffs with the care they give a pull request.
- Is Plannotator safe for company data?
- No telemetry. Content stays local except URL annotation (through Jina Reader by default), pull request fetches, AI questions sent to your chosen provider, and shares; PLANNOTATOR_SHARE=disabled turns sharing off.
- What are the best Plannotator alternatives?
- The closest alternatives we have reviewed are Vibe Kanban and Claude Squad.
Answers checked Nov 14, 2026.