Chrome DevTools Protocol MCP server for browser debugging with tools for console, network, screenshots, and DOM interaction.
SaferSkills independently audited argus (MCP Server) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
<p align="center"> <img src="src/assets/argus_banner.png" alt="Argus" width="1000" /> </p>
<p align="center"> Chrome DevTools Protocol MCP server — give AI agents eyes into a live browser. </p>
<p align="center"> <img src="https://img.shields.io/badge/MCP-compatible-blue?style=flat-square" alt="MCP compatible" /> <img src="https://img.shields.io/badge/node-%3E%3D18-brightgreen?style=flat-square" alt="Node ≥ 18" /> <img src="https://img.shields.io/badge/license-MIT-green?style=flat-square" alt="MIT" /> </p>
Argus connects directly to Chrome via the Chrome DevTools Protocol and exposes browser debugging as MCP tools - no Puppeteer, no Playwright, no browser abstraction layer. Spawn Chrome, attach to tabs, record console output, intercept network requests, inject mocks, and capture screenshots, all from your MCP client.
The name comes from Argus Panoptes - the 100-eyed giant of Greek mythology who could watch everything at once and never fully slept. That's the goal: total visibility into what the browser is doing.
Argus came about as part of an ongoing exploration into leveraging agents to reproduce reported issues as well as "manually" debug/record them on the fly. As a result you can expect to see bugs in Argus itself as well as plenty room for improvement and growth over time.
console.log/warn/error and unhandled exceptions with stack traces# Install globally
npm install -g @jmsa/argus-mcp
argus-mcp
# Or run without installing
npx @jmsa/argus-mcpgit clone https://github.com/Jmsa/argus
cd argus
npm install
npm run devChrome Canary opens automatically on startup with the Argus welcome page. Connect your MCP client to the stdio transport and start using the tools.
Claude Code — run this once:
# npm package (recommended)
claude mcp add --transport stdio argus -- argus-mcp
# from source
claude mcp add --transport stdio argus -- npm run devClaude Desktop — add to claude_desktop_config.json:
{
"mcpServers": {
"argus": {
"type": "stdio",
"command": "argus-mcp"
}
}
}If running from source instead, use"command": "npm"with"args": ["run", "dev"]and set"cwd"to the project root.
By default, Argus does not automatically launch Chrome when a Claude session starts. Use the browser_launch tool to start Chrome when you need it.
The first time Claude starts with Argus installed, Chrome opens once automatically so you can verify everything is working. A message on the welcome page explains what's happening. After that first session, auto-launch is off unless you enable it.
The welcome page always shows a Yes / No toggle for auto-launch:
browser_launch manuallyThe toggle takes effect immediately and persists across sessions. Your preference is stored in ~/.argus/config.json and can also be edited manually:
{ "autoLaunch": true }Set ARGUS_NO_LAUNCH=1 to force-skip Chrome launch regardless of the config file. This is useful in CI pipelines or automated environments where a display isn't available.
Argus exposes 32 tools across eight groups.
| Tool | Description |
|---|---|
browser_launch | Spawn a new Chrome instance with remote debugging |
browser_connect | Attach to an already-running Chrome via WebSocket URL |
browser_disconnect | Disconnect (browser stays open) |
browser_status | Check connection state and active tab count |
| Tool | Description |
|---|---|
tab_list | List all open page tabs |
tab_open | Open a new tab and navigate to a URL |
tab_navigate | Navigate an existing tab to a new URL |
tab_close | Close a tab by targetId |
tab_screenshot | Capture a screenshot (viewport, full-page, or clipped) |
| Tool | Description |
|---|---|
console_start | Begin recording console output and exceptions |
console_stop | Stop recording |
console_get_logs | Retrieve logs, filterable by type and text |
console_clear | Discard captured log entries |
| Tool | Description |
|---|---|
network_start_recording | Enable network capture (requests, responses, bodies) |
network_stop_recording | Disable network capture |
network_get_requests | Query captured requests (filter by URL, method, status, error) |
network_clear_requests | Clear the request history for a tab |
| Tool | Description |
|---|---|
network_add_mock | Intercept requests matching a glob and return a custom response |
network_remove_mock | Remove a mock rule by ID |
network_list_mocks | List active mock rules for a tab |
network_clear_mocks | Remove all mocks and disable interception |
| Tool | Description |
|---|---|
page_evaluate | Execute JavaScript and return the result |
page_reload | Reload the tab (optionally bypassing cache) |
page_get_url | Get the current URL and title of a tab |
| Tool | Description |
|---|---|
dom_query | Query the first element matching a CSS selector and return its properties |
dom_query_all | Query all elements matching a CSS selector |
dom_click | Click the first element matching a CSS selector (scrolls into view first) |
dom_input_value | Set an input's value and dispatch input/change events (React/Vue safe) |
dom_get_value | Get the current value of an input element |
dom_wait_for | Wait for an element to appear in the DOM (polls every 100ms) |
| Tool | Description |
|---|---|
banner_update | Push state updates to the Argus banner overlay (recording indicator, counts) |
banner_get_screenshots | Retrieve screenshots captured via the banner Screenshot button |
Skills are Claude Code workflows that invoke Argus tools automatically. Install the plugin to get them as slash commands:
/plugin install Jmsa/argus| Skill | Command | Description |
|---|---|---|
debug-session | /argus:debug-session <url> | Capture a complete debugging snapshot — console, network, screenshot |
repro-issue | /argus:repro-issue <url> <bug description> | Reproduce a bug using mocks to isolate frontend vs API |
network-debug | /argus:network-debug <url> | Investigate failed requests, slow responses, and mock verification |
MCP Client (Claude, Inspector, etc.)
│ stdio
▼
Argus MCP Server
│ CDP over WebSocket
▼
Chrome / Chrome Canary
│ per-tab CDPSession
▼
domains: console · screenshot · network · uiChrome is spawned as a child process. Argus listens to its stderr for the DevTools listening on ws://... line to get the exact WebSocket URL, then connects. Each tab gets its own CDPSession (multiplexed over a single WebSocket connection) with independent domain state.
npm run dev # start with tsx (no build step)
npm run build # compile to dist/
npm run typecheck # type-check without emittingChrome profile data is stored at ~/.argus/chrome-profile so Chrome doesn't reinitialise on every restart.
Contributions are welcome. Argus is an active exploration project, so there's plenty of room to improve.
Before opening a PR:
mainnpm run typecheck — PRs must pass type checkingGood areas to contribute:
Open an issue first for large changes so we can align on approach before you invest time in implementation.
MIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.