AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
MCP verified MIT Self-run

Webclaw

mcp-kuroko1t-webclaw · by kuroko1t

A WebMCP-native browser agent that runs inside your real Chrome — control it from Claude, Cursor, and any MCP client

No reviews yet
0 installs
37 views
0.0% view→install

Install

$ agentstack add mcp-kuroko1t-webclaw

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access Used
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/mcp-kuroko1t-webclaw)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
6mo ago

Declared compatibility

Claude CodeClaude DesktopCursorWindsurf

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Webclaw? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

WebClaw

[](https://github.com/kuroko1t/webclaw/actions/workflows/ci.yml) [](https://www.npmjs.com/package/webclaw-mcp) [](LICENSE)

A WebMCP-native browser agent that runs inside your real Chrome — control it from Claude, Cursor, and any MCP client.

Why WebClaw?

| | WebClaw | browser-use | Playwright MCP | |---|---|---|---| | Runs in real Chrome | Yes (extension) | No (CDP) | No (Playwright) | | User's logged-in sessions | Yes | No | No | | Bot detection | Resistant | Vulnerable | Vulnerable | | WebMCP native tools | Yes | No | No | | Page understanding | @ref a11y tree | Screenshots | DOM / Screenshots |

WebClaw runs as a Chrome extension, so it sees exactly what you see — cookies, logins, extensions, and all. No headless browser, no CDP injection, no bot flags.

Quick Start

1. Configure your MCP client

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "webclaw": {
      "command": "npx",
      "args": ["-y", "webclaw-mcp"]
    }
  }
}

Config file locations:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Linux: ~/.config/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Claude Code

claude mcp add webclaw -- npx -y webclaw-mcp

Cursor

Add to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "webclaw": {
      "command": "npx",
      "args": ["-y", "webclaw-mcp"]
    }
  }
}

VS Code (Copilot)

Add to .vscode/mcp.json in your project root:

{
  "servers": {
    "webclaw": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "webclaw-mcp"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "webclaw": {
      "command": "npx",
      "args": ["-y", "webclaw-mcp"]
    }
  }
}

2. Load the Chrome extension

  1. Download the latest webclaw-extension.zip from the Releases page and unzip
  2. Open chrome://extensions/ → enable Developer mode
  3. Click Load unpacked → select the dist/ folder

Or build from source

git clone https://github.com/kuroko1t/webclaw.git && cd webclaw && pnpm install && pnpm build

Then load packages/extension/dist/ as above.

3. Try it

Restart your MCP client and ask:

> "Go to wikipedia.org and search for Model Context Protocol"

Chrome launches automatically on the first tool call. The Side Panel shows all activity in real-time.

How It Works

flowchart LR
    LLM["AI Assistant"]
    MCP["MCP Server"]
    EXT["Chrome Extension"]
    CS["Content Script"]

    LLM -- "stdio" --> MCP -- "WebSocket" --> EXT --> CS

    style LLM fill:#7b2d8b,stroke:#333,color:#fff
    style MCP fill:#1a5276,stroke:#333,color:#fff
    style EXT fill:#1e8449,stroke:#333,color:#fff
    style CS fill:#b7950b,stroke:#333,color:#fff
  1. The AI calls MCP tools (navigate_to, page_snapshot, click, ...)
  2. The MCP server forwards requests to the Chrome extension via WebSocket
  3. The content script reads the page as a compact accessibility tree with @ref labels
  4. The AI picks the right @ref and acts — no pixel coordinates, no CSS selectors

page_snapshot output

[page "Wikipedia, the free encyclopedia"]
  [search]
    [form]
      [@e3 searchbox "Search Wikipedia"]    ← AI targets this
      [@e4 button "Search"]                 ← and clicks this
  [main]
    [heading[1] "Welcome to Wikipedia"]
    [group]
      [heading[2] "From today's featured article"]
      ...

Every interactive element gets a stable @ref label. The AI reads the tree, picks the right @ref, and calls click / type_text / select_option.

MCP Tools (21)

Page Interaction — navigate, snapshot, click, hover, type, select, drop files, screenshot, dialog, evaluate

| Tool | Parameters | Description | |------|-----------|-------------| | navigate_to | url, tabId? | Navigate to a URL | | page_snapshot | focusRegion?, interactiveOnly?, tabId? | Get a compact accessibility tree with @ref labels | | click | ref, snapshotId, tabId? | Click an element by its @ref label | | hover | ref, snapshotId, tabId? | Hover over an element to reveal hidden UI (dropdowns, tooltips) | | type_text | ref, text, snapshotId, clearFirst?, tabId? | Type text into an input/textarea by @ref | | select_option | ref, value, snapshotId, tabId? | Select a dropdown option by @ref | | drop_files | ref, snapshotId, files, tabId? | Drop files onto an element (e.g., upload images to GitHub issues) | | screenshot | tabId? | Capture the visible area of the active tab | | handle_dialog | action, promptText?, tabId? | Handle a native browser dialog (alert/confirm/prompt) | | evaluate | expression, tabId? | Evaluate JavaScript in the page context (e.g., localStorage, cookies, computed styles) |

Tab Management — new, list, switch, close

| Tool | Parameters | Description | |------|-----------|-------------| | new_tab | url? | Open a new tab (optionally with a URL) | | list_tabs | | List all open tabs with tabId, URL, title, and active state | | switch_tab | tabId | Switch to a specific tab | | close_tab | tabId | Close a specific tab |

Navigation — back, forward, reload, wait, scroll

| Tool | Parameters | Description | |------|-----------|-------------| | go_back | tabId? | Go back to the previous page | | go_forward | tabId? | Go forward to the next page | | reload | tabId?, bypassCache? | Reload the current page | | wait_for_navigation | tabId?, timeoutMs? | Wait for the page to finish loading | | scroll_page | direction?, amount?, ref?, snapshotId?, tabId? | Scroll the page or scroll to a specific element |

WebMCP — discover and invoke page-declared tools

| Tool | Parameters | Description | |------|-----------|-------------| | list_webmcp_tools | tabId? | Discover tools on the page (native WebMCP + auto-synthesized from buttons/forms) | | invoke_webmcp_tool | toolName, args?, tabId? | Invoke a discovered tool (native or synthesized) |

Development

git clone https://github.com/kuroko1t/webclaw.git
cd webclaw
pnpm install
pnpm build        # Build all packages
pnpm test         # Run all tests
pnpm dev          # Watch mode

Project Structure

packages/
  shared/          Type definitions, Zod schemas, utilities
  mcp-server/      MCP server with 21 tools, WebSocket bridge
  extension/       Chrome MV3 extension (service worker, content scripts, side panel)
examples/
  webmcp-demo-site/  WebMCP-enabled Todo app for testing native tool discovery

Multi-Session Support

WebClaw automatically scans ports 18080–18089, so you can run up to 10 MCP server instances simultaneously (e.g., multiple Claude Code sessions). Each instance binds to the first available port, and the Chrome extension connects to all of them.

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | WEBCLAW_PORT | 18080 | WebSocket port for MCP server ↔ extension communication. When set, disables auto-scanning and uses only this port. |

Troubleshooting

Chrome extension not connected

  • Chrome is auto-launched when a tool is called. If it fails, start Chrome manually with the extension loaded
  • Check the Service Worker console (chrome://extensions/ → Details → Service Worker) for Connected to MCP server
  • Verify the MCP server is running (look for WebSocket server listening on 127.0.0.1: in stderr)

MCP client cannot connect

  • Ensure npx webclaw-mcp runs successfully from your terminal
  • Check for port conflicts: WebClaw auto-scans 18080–18089. If all are in use, set WEBCLAW_PORT to a custom port
  • Restart your MCP client after updating the config

Content script not injecting

  • Content scripts only inject into http:// and https:// pages (not chrome://, file://, etc.)
  • Try refreshing the page after loading the extension

WebMCP tools not discovered

  • Native WebMCP requires Chrome 146+ with the navigator.modelContext API
  • Verify the page declares tools via navigator.modelContext.addTool() or ``

Contributing

Contributions are welcome! Please open an issue to discuss your idea before submitting a pull request.

Disclaimer

WebClaw is a browser automation tool intended for legitimate use cases such as personal productivity, development, testing, and accessibility. Users are responsible for complying with the terms of service of any website they interact with. The authors are not responsible for any misuse of this tool.

License

[MIT](LICENSE)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.