claude-annotate
Enables visual annotation on web pages for Claude Code, allowing element selection, comment addition, screenshot capture, and structured UI feedback for code fixes via an MCP server.
README
<p> <img src="banner.png" alt="Claude Annotate" width="1100"> </p>
Claude Annotate
Visual annotation for AI. Click elements, capture screenshots, fix code.
/annotate
Figma-like annotation experience with floating inline note cards. DevTools-like element picker in vanilla JS.
Click elements, add comments, submit. The agent gets selectors, box model, accessibility, screenshots — everything it needs to fix your UI.
https://github.com/user-attachments/assets/115b10ca-86e8-4b1c-b8a4-492c68759c58
Fork note — Claude Code support. This fork adds an MCP server in
server/that brings the same annotation workflow to Claude Code. The Chrome extension and native host are reused unchanged — only the agent layer differs. Upstream: nicobailon/claude-annotate.
Claude Code (MCP)
cd server
npm install && npm run build
# Load chrome-extension/ in Chrome (Developer Mode), note the extension ID, then:
( cd ../chrome-extension/native && ./install.sh <extension-id> ) # restart browser
claude mcp add claude-annotate -- node "$PWD/dist/index.js" # restart Claude Code
Then ask Claude to annotate a page — it calls the annotate tool, you click/comment in the browser, and Claude receives selectors, box model, accessibility, your comments, and screenshots (returned both inline and as temp files). Default wait is 600s. Full setup, config, and tests: server/README.md.
Quick Start (Pi)
1. Install Pi Extension
pi install npm:claude-annotate
Restart pi to load the extension.
2. Load Supported Browser Extension
- Open the extensions page in Google Chrome, Google Chrome for Testing, or Chromium, and enable Developer mode
- Click Load unpacked → select the
chrome-extension/folder inside the installed package - Click the Claude Annotate icon in the toolbar
3. Install Native Host
The popup shows your extension ID. Click Copy next to the install command, then run it from chrome-extension/native/ in the installed package:
./install.sh <extension-id>
This installs the native messaging manifest for Google Chrome, Google Chrome for Testing, and Chromium on macOS, plus the default/current config-home locations for those browsers on Linux. Fully quit and reopen that browser. The popup will show Connected when ready.
Usage
/annotate # Current browser tab
/annotate https://x.com # Opens URL first
| Action | How |
|---|---|
| Select element | Click on page |
| Cycle ancestors | Alt/⌥+scroll while hovering |
| Multi-select | Toggle "Multi" or Shift+click |
| Add comment | Type in note card textarea |
| Toggle screenshot | 📷 button in note card header |
| Reposition note | Drag by header |
| Scroll to element | Click selector in note card |
| Toggle note | Click numbered badge |
| Expand/collapse all | ▼/▲ buttons in toolbar |
| Toggle edit capture | "Etch" toggle in toolbar |
| Toggle annotation UI | ⌘/Ctrl+Shift+P |
| Close | ESC |
Features
Context Capture — Each element automatically gets box model breakdown (padding, border, margin), accessibility info (role, name, focusable, ARIA states), all HTML attributes, and key CSS styles (display, position, overflow, colors, typography). Enable Debug mode for computed styles (40+ properties), parent context, and CSS variables.
Inline Note Cards — Draggable floating cards with per-element comments, SVG connectors linking notes to elements, click-to-scroll, and per-element screenshot toggles.
Screenshots — Individual crops per element (20px padding) or full-page mode with numbered badges drawn on the screenshot to identify elements. Toggle per element with the 📷 button.
Edit Capture — Toggle "Etch" in the toolbar to record DevTools edits. Change inline styles, modify CSS rules, add/remove classes, edit text — everything is tracked via MutationObserver. A pulsing red dot and badge counter show recording status. At submit, the extension takes before/after screenshots by briefly undoing visual changes, and produces structured property-level diffs the agent can map to source code. Works alongside element selection or standalone.
Restricted Tabs — If the current tab is chrome:// or other restricted URLs, providing a URL opens a new tab automatically. Popup button and keyboard shortcut auto-inject the content script on fresh tabs.
Output
## Page Annotation: https://example.com
**Viewport:** 1440×900
**Context:** Fix the styling issues
### Selected Elements (2)
1. **button**
- Selector: `#submit-btn`
- ID: `submit-btn`
- Classes: `btn, btn-primary`
- Text: "Submit"
- **Box Model:** 120×40 (content: 96×24, padding: 8 16, border: 1, margin: 0 8)
- **Attributes:** type="submit", data-testid="submit"
- **Styles:** display: flex, backgroundColor: rgb(59, 130, 246)
- **Accessibility:** role=button, name="Submit", focusable=true, disabled=false
- **Comment:** Make this blue with rounded corners
2. **div**
- Selector: `.error-message`
- Classes: `error-message, hidden`
- Text: "Please fill required fields"
- **Box Model:** 300×20 (content: 300×20, padding: 0, border: 0, margin: 0 0 8)
- **Accessibility:** focusable=false, disabled=false
- **Comment:** This should appear in red, not hidden
### Screenshots
- Element 1: /var/folders/.../claude-annotate-...-el1.png
- Element 2: /var/folders/.../claude-annotate-...-el2.png
## Edit Capture (2 changes, 35s)
### Inline Style Changes
**`#submit-btn`**
- `background-color`: `rgb(59, 130, 246)` → `rgb(37, 99, 235)`
- `border-radius`: added `8px`
### CSS Rule Changes
**`.btn-primary:hover`** (styles.css)
- `background-color`: `rgb(37, 99, 235)` → `rgb(29, 78, 216)`
### Before/After Screenshots
- Before: /var/folders/.../claude-annotate-...-before.png
- After: /var/folders/.../claude-annotate-...-after.png
Debug mode adds computed styles, parent context, and CSS variables per element. Edit capture appears when the Etch toggle is enabled and changes are detected.
Architecture
Pi Extension (index.ts)
↕ Unix Socket (/tmp/claude-annotate.sock)
Native Host (host.cjs)
↕ Browser Native Messaging
Browser Extension (background.js → content.js)
| File | Purpose |
|---|---|
index.ts |
Pi extension — /annotate command + tool |
types.ts |
TypeScript interfaces |
chrome-extension/content.js |
Element picker UI (vanilla JS) |
chrome-extension/background.js |
Native messaging, screenshots, tab routing |
chrome-extension/native/host.cjs |
Socket ↔ native messaging bridge |
chrome-extension/popup.html |
Connection status + setup |
Auth token generated per-run at /tmp/claude-annotate.token. Socket and token files use 0600 permissions.
Claude Code variant: the server/ MCP server replaces the Pi extension at the top of this stack — it speaks the same Unix-socket protocol to the unchanged native host, so the bottom two layers are identical. See server/README.md.
Development
No build step. Edit content.js or background.js directly, reload at chrome://extensions. Pi extension (TypeScript) loads via jiti — restart pi after changes.
tail -f /tmp/claude-annotate-host.log # Native host logs
# chrome://extensions → Claude Annotate → service worker # Background logs
# DevTools on target page # Content script logs
Troubleshooting
| Issue | Fix |
|---|---|
| UI doesn't appear | Refresh page, check service worker console |
| "restricted URL" error | Provide a URL: /annotate https://example.com |
| Native host not connecting | Click extension icon → check status, re-run install, fully restart the supported browser |
| "Extension ID mismatch" | Copy install command from popup, re-run |
| Socket errors | ls -la /tmp/claude-annotate.sock |
Verify native host:
- macOS Google Chrome:
cat ~/Library/Application\ Support/Google/Chrome/NativeMessagingHosts/com.claude.annotate.json - macOS Google Chrome for Testing:
cat ~/Library/Application\ Support/Google/ChromeForTesting/NativeMessagingHosts/com.claude.annotate.json - macOS Chromium:
cat ~/Library/Application\ Support/Chromium/NativeMessagingHosts/com.claude.annotate.json - Linux Google Chrome (default path):
cat ~/.config/google-chrome/NativeMessagingHosts/com.claude.annotate.json - Linux Google Chrome for Testing (default path):
cat ~/.config/google-chrome-for-testing/NativeMessagingHosts/com.claude.annotate.json - Linux Chromium (default path):
cat ~/.config/chromium/NativeMessagingHosts/com.claude.annotate.json - Linux with custom config home:
echo "${CHROME_CONFIG_HOME:-${XDG_CONFIG_HOME:-$HOME/.config}}"
If your Linux browser uses a different XDG config root, export CHROME_CONFIG_HOME or XDG_CONFIG_HOME before running ./install.sh <extension-id>. Custom --user-data-dir layouts are not handled by this installer.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.