Skip to main content
Glama

CDP MCP

Browser automation through Chrome DevTools Protocol. No Playwright. No Puppeteer. Just raw CDP over WebSocket.

Why

Playwright MCP runs headless — and many websites detect and block it. Chrome in Claude's computer use struggles with file uploads, making tasks like applying for jobs harder than they should be. Both are heavier and more brittle than they need to be.

CDP MCP talks directly to Chrome over DevTools Protocol, controls a real visible browser, reads the accessibility tree to discover every interactive element on the page, and verifies that interactions actually worked. More lightweight, more robust, and designed for AI agents that need to actually use the web.

Related MCP server: BrowserGenie MCP Server

Tools

Tool

What it does

cdp_launch

Launch Chrome with CDP enabled

cdp_connect

Connect to an existing Chrome instance

cdp_navigate

Go to URL, back, forward, refresh

cdp_snapshot

Get the accessibility tree with numbered [n] refs

cdp_interact

Click, type, select, check, press, upload, hover, focus, submit

cdp_execute

Run arbitrary JavaScript

cdp_wait

Wait for elements, text, navigation, network idle

cdp_screenshot

Capture viewport or full page

cdp_tabs

List, switch, close, or open tabs

cdp_cookies

Get, set, delete, clear cookies

cdp_console

Capture console output, JS errors, dialog info

How it works

The core loop is two tools: snapshot and interact.

cdp_snapshot returns the page's accessibility tree — a structured view of every interactive element, each tagged with a numbered ref:

[1] button "Sign In"
[2] textbox "Email" value=""
[3] textbox "Password" value=""
[4] checkbox "Remember me" checked=false
[5] link "Forgot password?"

cdp_interact accepts those [n] refs directly as selectors, along with CSS selectors and text= prefix matching:

cdp_interact(action="type", selector="[2]", value="user@example.com")
cdp_interact(action="click", selector="text=Sign In")
cdp_interact(action="click", selector="#submit-btn")

Interactions are verified — if a React controlled input silently rejects your value, you'll know:

{
  "success": true,
  "action": "type",
  "expected": "user@example.com",
  "actual": "user@example.com"
}

Framework handling

cdp_interact uses native property setters and dispatches the right events for React, Vue, and Angular inputs. It doesn't just set .value — it triggers the framework's internal change detection. If that fails, it falls back to CDP-level keystroke dispatch.

Installation

npm install
npm run build

Add to your MCP config:

{
  "mcpServers": {
    "cdp": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/cdp-mcp/dist/index.js"]
    }
  }
}

Test results

Tested against the-internet.herokuapp.com — 39 automation challenges covering forms, dynamic content, iframes, shadow DOM, drag and drop, and more. An AI agent (Claude) ran every test using only the tools above.

Test

Result

Form Authentication

Pass

Checkboxes

Pass

Dropdown

Pass

File Upload

Pass

JS Alert

Pass

JS Confirm

Pass

JS Prompt

Pass

Dynamic Controls

Pass

Hovers (CSS :hover)

Pass

WYSIWYG Editor (TinyMCE in iframe)

Pass

Sortable Data Tables

Pass

Key Presses

Pass

Shadow DOM

Pass

Context Menu (right-click)

Pass

Infinite Scroll

Pass

Dynamic Loading (render after)

Pass

Dynamic Loading (hidden reveal)

Pass

Nested Frames (frameset in frameset)

Pass

Add/Remove Elements

Pass

Disappearing Elements

Pass

Horizontal Slider

Pass

Inputs (number field)

Pass

Forgot Password (form submit)

Pass

Challenging DOM (unstable selectors)

Pass

Status Codes (404)

Pass

Redirect Link

Pass

Notification Messages (flash)

Pass

Typos (text detection)

Pass

Floating Menu

Pass

Broken Images (detection)

Pass

Drag and Drop (HTML5)

Pass

JQuery UI Menus (nested submenus)

Pass

Geolocation (API mocking)

Pass

A/B Testing

Pass

Entry Ad (modal dismiss)

Pass

Large & Deep DOM (182KB page)

Pass

Shifting Content

Pass

Multiple Windows

Pass

Dynamic Content

Pass

39/39.

Known code smells

  • cdp_tabs accesses CDPClient private fields (port, host) via as any cast. Works, needs proper getters.

Requirements

  • Node.js 18+

  • Chrome, Chromium, or Edge

Dependencies

  • ws — WebSocket client for CDP

  • That's it

License

Copyright (c) 2025 Cassius Oldenburg. All rights reserved.

Available Tools

27 tools
cdp_click_coordinatesA

Click at raw x,y screen coordinates. Useful for clicking elements in shadow DOM or other hard-to-select elements. Get coordinates from screenshots or cdp_find_elements rect values.

Inputs: x (required), y (required), click_count, button.

cdp_connectA

Connect to an already-running Chrome/Chromium instance with CDP enabled.

Inputs: port, host, tab.

cdp_epoch_infoB

Get current epoch info. Useful for debugging site memory behavior.

Inputs: no named parameters.

cdp_executeC

Execute JavaScript in the page context.

Inputs: script (required), frame_id.

cdp_find_elementsB

Discover interactive elements on the page (inputs, buttons, links, etc.).

Inputs: filter, selector, include_hidden, text_filter, +2 more.

cdp_fix_bugC

Mark a bug as fixed and remove it from the tracking file.

Inputs: id (required).

cdp_frame_interactB

Interact with elements inside iframes. Automatically searches all frames for the element.

Inputs: action (required), selector (required), value, frame_id.

cdp_interactC

Interact with elements: click, type, upload files, select options, etc.

Inputs: action (required), selector (required), value, file_path, +3 more.

cdp_launchA

Launch a Chrome/Chromium browser with CDP enabled. Creates a dedicated profile directory.

Inputs: port, headless, browser, profile, +3 more.

cdp_list_bugsB

List all tracked bugs in CDP-MCP.

Inputs: no named parameters.

cdp_list_framesA

List all frames on the page (main frame + iframes). Use this to find iframes for LinkedIn Easy Apply, Gmail compose, etc.

Inputs: no named parameters.

cdp_list_tabsB

List all open browser tabs.

Inputs: no named parameters.

cdp_monaco_editorA

Interact with Monaco Editor (VS Code web editor) on the page. Detects Monaco instances and provides reliable get/set operations that preserve formatting. Much better than generic text input tools for Monaco.

Inputs: action (required), value, editor_index.

cdp_navigateC

Navigate to a URL or perform navigation actions (back, forward, refresh).

Inputs: url, action, tab.

cdp_readC

Read content from the page: full page text, element text, attributes.

Inputs: target, selector, attribute.

cdp_screenshotB

Capture a screenshot of the page or a specific element.

Inputs: selector, format, quality, full_page, +1 more.

cdp_set_valueA

Set the value of an input/textarea, properly handling React/Vue/Angular controlled inputs. This clears existing content and sets new value, triggering all necessary events. Best for form fields that cdp_interact type struggles with.

Inputs: selector (required), value (required), clear_first.

cdp_site_clearB

Clear site memory for a domain (forces fresh scan next time).

Inputs: domain.

cdp_site_infoA

Get stored site memory for the current page or a specific domain. Returns iframe locations, shadow DOM elements, and interaction patterns learned from previous visits.

Inputs: domain.

cdp_site_listB

List all sites in memory with their key info.

Inputs: no named parameters.

cdp_site_noteB

Add a note or selector pattern to site memory. Use this to record what you learned about interacting with a site.

Inputs: note, selector_name, selector.

cdp_site_scanA

Scan current page structure and store in site memory. Detects iframes, shadow DOM, forms, and key elements. Automatically skips if already scanned this conversation epoch (compaction resets this). Use force=true to rescan anyway.

Inputs: force, note.

cdp_submit_formA

Submit a form reliably, with special handling for React/Vue/Angular. Tries multiple submission methods: requestSubmit(), clicking submit button, dispatching submit event, and direct submit(). Works when Enter key fails.

Inputs: selector.

cdp_track_bugA

Track a bug in the CDP-MCP. Writes to BUGS.md in the cdp-mcp folder. Use this when you encounter issues with browser automation.

Inputs: title (required), description (required), context.

cdp_type_textA

Type text at the current cursor position using CDP Input.insertText. This simulates real keyboard input and works with React/Vue/Angular controlled inputs. Focus an element first with cdp_interact focus action.

Inputs: text (required), delay, press_enter, press_tab.

cdp_upload_shadowA

Upload a file to a file input in shadow DOM. First use cdp_execute to find the file input and store it in window.__fileInput, then call this tool.

Inputs: file_path (required).

cdp_waitC

Wait for conditions: element visible, text contains, navigation, etc.

Inputs: condition (required), selector, value, timeout.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 27 tools

    Disambiguation3/5

    The tool set has clear distinctions for core browser automation tasks like navigation, interaction, and reading, but there is notable overlap between cdp_interact, cdp_set_value, and cdp_type_text for input handling, and between cdp_find_elements and cdp_site_scan for element discovery. Descriptions help clarify, but an agent might struggle to choose the optimal tool in some scenarios.

    Naming Consistency5/5

    All tools follow a consistent 'cdp_' prefix with snake_case naming, using clear verb_noun patterns (e.g., cdp_launch, cdp_navigate, cdp_read). This uniformity makes the tool set predictable and easy to navigate, with no deviations in style.

    Tool Count2/5

    With 27 tools, the count is excessive for a browser automation server, leading to potential confusion and redundancy. While the domain is broad, many tools (e.g., cdp_click_coordinates vs. cdp_interact for clicking) could be consolidated, making the set feel heavy and over-engineered.

    Completeness5/5

    The tool set comprehensively covers browser automation and site interaction, including connection, navigation, element discovery, interaction, reading, screenshot, form handling, and memory management. There are no obvious gaps; it supports full workflows from launch to debugging, with specialized tools for edge cases like shadow DOM and Monaco editor.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      An MCP server for stealth browser automation that uses human-like interaction patterns to bypass bot detection via the Chrome DevTools Protocol. It enables users to navigate, interact with elements, and capture data from websites using undetectable behaviors like Bezier mouse movements and Gaussian typing delays.
      2
      MIT
    • A
      license
      B
      quality
      D
      maintenance
      An MCP server that provides AI models with full browser automation capabilities through Chrome. It enables navigation, interaction, screenshots, and complete DevTools access by bridging AI clients with a companion Chrome extension.
      99
      3 npm
      3
      Apache 2.0