npx skills add ...
npx skills add reason-machines/devtools-skills --skill chrome-devtools-axi-browser-automation
Agent-ergonomic browser automation with token-efficient accessibility tree snapshots and combined operations
npx skills add reason-machines/devtools-skills --skill chrome-devtools-axi-browser-automation
Skill by ara.so — Devtools Skills collection.
chrome-devtools-axi is an AXI-compliant CLI wrapper around chrome-devtools-mcp that provides token-efficient browser automation. It uses TOON-encoded output to reduce token usage by ~40% compared to raw JSON, combines operations (navigate + snapshot + suggestions in one command), and includes contextual next-step hints in every response.
Key features:
uid= refs for precise element targetingFor agents to install automatically:
For manual installation:
Tell your agent:
The CLI communicates with a persistent bridge server (localhost:9224 by default) that maintains an MCP session with Chrome via DevTools Protocol. The bridge auto-starts on first command and writes state to ~/.chrome-devtools-axi/.
Elements in snapshots are marked with uid=g<generation>:<id> refs. Always use refs exactly as printed:
Available types: log, debug, info, error, warn, dir, dirxml, table, trace, clear, assert, all
Available types: document, stylesheet, image, media, font, script, xhr, fetch, websocket, manifest, other, all
State stored in ~/.chrome-devtools-axi/:
bridge.pid — Running bridge PID and portsnapshot-generation — Staleness detection counterRefs use a generation prefix (g<N>:) that increments with each snapshot. If you capture a snapshot at generation 1, get ref @g1:5, then the page re-renders and moves to generation 2, attempting to use @g1:5 will fail with STALE_REF error.
Agent pattern:
STALE_REF error, re-snapshot and retry with new refAlways use refs exactly as printed in snapshot output. If you get STALE_REF, capture a new snapshot.
If uid is valid but element doesn't respond:
chrome-devtools-axi wait before interactionopen gives snapshot automatically, reducing round-trips--full sparingly — Default truncation saves tokens; only expand when neededfillform over multiple fill — Batch form operations when possiblewait <text> instead of arbitrary timeouts when possible# Click an element
chrome-devtools-axi click @g1:1
# Fill a form field
chrome-devtools-axi fill @g2:5 "user@example.com"
# Fill multiple fields at once
chrome-devtools-axi fillform @g2:5=john @g2:6=doe @g2:7=john@example.com
# Type text at current focus
chrome-devtools-axi type "search query"
# Press keyboard key
chrome-devtools-axi press Enter
chrome-devtools-axi press Tab
# Hover over element
chrome-devtools-axi hover @g1:3
# Drag and drop
chrome-devtools-axi drag @g1:4 @g1:8
# Upload file
chrome-devtools-axi upload @g1:2 /path/to/file.pdf
# Handle browser dialogs
chrome-devtools-axi dialog accept
chrome-devtools-axi dialog dismiss# Save screenshot to file
chrome-devtools-axi screenshot output.png
# Capture specific element
chrome-devtools-axi screenshot element.png --uid @g1:5
# Full-page screenshot
chrome-devtools-axi screenshot full.png --full-page
# Specify format
chrome-devtools-axi screenshot output.jpg --format jpeg# Evaluate expression (auto-wrapped)
chrome-devtools-axi eval "document.title"
# Execute function
chrome-devtools-axi eval "() => document.querySelectorAll('a').length"
# Multi-statement logic
chrome-devtools-axi eval "() => {
const rows = [...document.querySelectorAll('tr')];
return rows.map(row => row.textContent.trim());
}"# Execute script from stdin
cat << 'EOF' | chrome-devtools-axi run
open https://example.com
wait "Example Domain"
click @g1:1
snapshot
EOF# List all tabs
chrome-devtools-axi pages
# Open new tab
chrome-devtools-axi newpage https://google.com
# Open in background
chrome-devtools-axi newpage https://github.com --background
# Switch tab
chrome-devtools-axi selectpage <id>
# Close tab
chrome-devtools-axi closepage <id>
# Resize viewport
chrome-devtools-axi resize 1920 1080# Emulate mobile viewport
chrome-devtools-axi emulate --viewport "390x844x3,mobile"
# Set color scheme
chrome-devtools-axi emulate --color-scheme dark
# Network throttling
chrome-devtools-axi emulate --network "Slow 3G"
# CPU throttling (1-20)
chrome-devtools-axi emulate --cpu 4
# Set geolocation
chrome-devtools-axi emulate --geolocation "37.7749x-122.4194"
# Custom user agent
chrome-devtools-axi emulate --user-agent "CustomBot/1.0"
# Combine options
chrome-devtools-axi emulate --viewport "390x844x3,mobile" --color-scheme dark --network "Fast 3G"# List all console messages
chrome-devtools-axi console
# Filter by type
chrome-devtools-axi console --type error
chrome-devtools-axi console --type warn
# Pagination
chrome-devtools-axi console --limit 10 --page 2
# Get specific message
chrome-devtools-axi console-get <id># List network requests
chrome-devtools-axi network
# Filter by type
chrome-devtools-axi network --type xhr
chrome-devtools-axi network --type fetch
# Get specific request
chrome-devtools-axi network-get <id>
# Save response body
chrome-devtools-axi network-get <id> --response-file response.json
# Save request body
chrome-devtools-axi network-get <id> --request-file request.json# Run Lighthouse audit
chrome-devtools-axi lighthouse
# Mobile device
chrome-devtools-axi lighthouse --device mobile
# Snapshot mode (current state)
chrome-devtools-axi lighthouse --mode snapshot
# Save reports to directory
chrome-devtools-axi lighthouse --output-dir ./reports# Start trace
chrome-devtools-axi perf-start
# Start without reload
chrome-devtools-axi perf-start --no-reload
# Stop and save trace
chrome-devtools-axi perf-stop --file trace.json
# Analyze performance insight
chrome-devtools-axi perf-insight lcp "Largest Contentful Paint"# Capture heap snapshot
chrome-devtools-axi heap memory.heapsnapshot# Change bridge server port (default: 9224)
export CHROME_DEVTOOLS_AXI_PORT=9225
# Connect to existing Chrome instance (HTTP/HTTPS)
export CHROME_DEVTOOLS_AXI_BROWSER_URL=http://127.0.0.1:9222
# Connect via WebSocket directly
export CHROME_DEVTOOLS_AXI_BROWSER_URL=wss://cluster.example/launch
# WebSocket authentication headers (JSON)
export CHROME_DEVTOOLS_AXI_WS_HEADERS='{"Authorization":"Bearer token"}'
# Disable session hook auto-install
export CHROME_DEVTOOLS_AXI_DISABLE_HOOKS=1# Navigate and fill login form
chrome-devtools-axi open https://app.example.com/login
chrome-devtools-axi fillform @g1:1=user@example.com @g1:2=password
chrome-devtools-axi click @g1:3
chrome-devtools-axi wait "Dashboard"
chrome-devtools-axi snapshot# Extract form data with JavaScript
chrome-devtools-axi eval "() => {
const form = document.querySelector('form');
const data = new FormData(form);
return Object.fromEntries(data.entries());
}"// In TypeScript/Node.js
import { execSync } from 'child_process';
function runCommand(cmd: string): string {
return execSync(`chrome-devtools-axi ${cmd}`, { encoding: 'utf-8' });
}
// Navigate and extract data
runCommand('open https://example.com');
const title = runCommand('eval "document.title"');
// Take screenshot
runCommand('screenshot page.png');
// Fill and submit form
runCommand('fill @g1:1 "search term"');
runCommand('press Enter');
runCommand('wait 2000');
// Extract results
const results = runCommand('eval "() => [...document.querySelectorAll(\'.result\')].map(el => el.textContent)"');# Wait for element to appear, then interact
chrome-devtools-axi open https://spa.example.com
chrome-devtools-axi wait "Loading complete"
chrome-devtools-axi snapshot
chrome-devtools-axi click @g2:5# Test mobile layout
chrome-devtools-axi emulate --viewport "390x844x3,mobile"
chrome-devtools-axi open https://example.com
chrome-devtools-axi screenshot mobile.png
# Test desktop layout
chrome-devtools-axi resize 1920 1080
chrome-devtools-axi snapshot
chrome-devtools-axi screenshot desktop.png# First snapshot
chrome-devtools-axi snapshot
# Output shows: uid=g1:1 link "Click here"
# Page changes, generation bumps to g2
chrome-devtools-axi click @g1:1
# Error: STALE_REF
# Re-snapshot to get fresh refs
chrome-devtools-axi snapshot
# Output shows: uid=g2:3 link "Click here"
chrome-devtools-axi click @g2:3
# Success# Manually stop and restart bridge
chrome-devtools-axi stop
chrome-devtools-axi start# Use different port
export CHROME_DEVTOOLS_AXI_PORT=9999
chrome-devtools-axi open https://example.com# Disable truncation for complete data
chrome-devtools-axi snapshot --full
chrome-devtools-axi console --full# Check for errors in console
chrome-devtools-axi console --type error
# Test simple eval first
chrome-devtools-axi eval "2 + 2"
# Use arrow function for complex logic
chrome-devtools-axi eval "() => { console.log('debug'); return 42; }"#!/bin/bash
# Start fresh session
chrome-devtools-axi stop
chrome-devtools-axi start
# Navigate to app
chrome-devtools-axi open https://app.example.com
# Login
chrome-devtools-axi fillform @g1:1=$USER_EMAIL @g1:2=$USER_PASSWORD
chrome-devtools-axi click @g1:3
chrome-devtools-axi wait "Dashboard"
# Take baseline screenshot
chrome-devtools-axi screenshot baseline.png
# Navigate to feature
chrome-devtools-axi click @g2:10
chrome-devtools-axi wait "Feature Page"
# Interact with feature
chrome-devtools-axi fill @g3:5 "test data"
chrome-devtools-axi click @g3:6
# Verify result
chrome-devtools-axi wait "Success"
chrome-devtools-axi snapshot --full > result.txt
# Capture performance
chrome-devtools-axi lighthouse --output-dir ./reports
# Cleanup
chrome-devtools-axi stop