Selenix-MCP
The Selenix-MCP server connects Claude Desktop to the Selenix browser automation platform, enabling AI-assisted test creation, execution, management, and debugging through natural language.
Reading & Inspection
get_test_list– List all tests with IDs, names, and command countsget_current_test/get_test_by_name– Retrieve a test and its commandsget_project_info– Project metadata (name, base URL, test/suite counts)get_active_suite– Info about the currently active test suiteget_workspace_context– High-level summary of the current workspaceget_command_list/get_command_info– List available Selenix commands or get detailed docs for a specific oneget_step_results– Pass/fail/error results from the last test runget_logs– Paginated recent logs for monitoring and troubleshooting
Browser State Capture
get_screenshot– Capture a screenshot of the active browser windowget_page_html– Get annotated HTML of the current page (scripts/styles removed)get_search_html– Search page HTML using a regex pattern with surrounding contextget_selected_command_info– Detailed DOM hierarchy, accessibility info, and element context for a command's target
Test Creation & Modification
add_commands– Insert one or more commands into a test at a specific positionclear_and_replace_commands– Completely replace all commands in a testfix_commands– Apply targeted fixes by updating, removing, or inserting commands at specific indexes
Execution
run_test– Execute a test and receive per-command pass/fail results
Selenix MCP Server
Connect Claude Desktop to Selenix via the Model Context Protocol. Chat with Claude to create, run, debug, and manage browser automation tests — just like using the built-in AI Assistant, but powered by Claude.
Prerequisites
Selenix desktop app installed
Claude Desktop installed
Node.js 18 or later
Related MCP server: playwright-mcp
Enable the Bridge in Selenix
Before using any setup method below, enable the bridge inside Selenix:
Open Selenix
Go to System Settings (gear icon in the sidebar)
Scroll to the MCP Server section
Set Enable MCP Server to Yes
This starts a local bridge server that the MCP server connects to.
Setup
Choose one of the following methods:
Option A: Install via npm (recommended)
npm install -g @selenix/mcp-serverThen find your global npm path:
npm root -gOpen your Claude Desktop config file:
Windows: Press
Win+R, paste%APPDATA%\Claude\claude_desktop_config.json, press EntermacOS: Open
~/Library/Application Support/Claude/claude_desktop_config.json
Add the Selenix MCP server, replacing the path with the output from npm root -g:
{
"mcpServers": {
"selenix": {
"command": "node",
"args": ["<npm-root-path>/@selenix/mcp-server/dist/bundle.js"]
}
}
}Example (Windows with nvm):
{
"mcpServers": {
"selenix": {
"command": "node",
"args": ["C:/Users/YourName/AppData/Local/nvm/v18.20.8/node_modules/@selenix/mcp-server/dist/bundle.js"]
}
}
}Example (macOS/Linux):
{
"mcpServers": {
"selenix": {
"command": "node",
"args": ["/usr/local/lib/node_modules/@selenix/mcp-server/dist/bundle.js"]
}
}
}If you already have other MCP servers configured, add "selenix" alongside them inside the existing "mcpServers" object.
Option B: Use the bundled version (no npm required)
If you installed Selenix as a desktop app, the MCP server is already bundled. No need to install anything — just point Claude Desktop to the bundled file.
Find your Selenix install folder and look for resources/mcp-server/index.js, then add to your Claude Desktop config:
{
"mcpServers": {
"selenix": {
"command": "node",
"args": ["C:/path/to/Selenix/resources/mcp-server/index.js"]
}
}
}Typical Windows path:
C:/Users/YourName/AppData/Local/Programs/Selenix/resources/mcp-server/index.jsOption C: Manual download
Download
bundle.jsfrom the npm package page or from the releases pageSave it somewhere on your machine (e.g.,
C:/selenix-mcp/bundle.js)Add to your Claude Desktop config:
{
"mcpServers": {
"selenix": {
"command": "node",
"args": ["C:/selenix-mcp/bundle.js"]
}
}
}Node.js is still required — the MCP server is a single JavaScript file with no other dependencies.
Verify
Restart Claude Desktop after editing the config. You should see the Selenix tools icon (hammer) in the chat input area.
Try asking Claude:
"What tests are in my Selenix project?"
"Create a test that opens google.com and searches for 'selenium automation'"
"Run the current test and fix any failures"
"Take a screenshot of the browser"
How It Works
Claude Desktop <--stdio--> MCP Server <--HTTP--> Bridge Server (inside Selenix)The MCP server translates Claude's tool calls into HTTP requests to a bridge server running inside Selenix. The bridge has full access to Selenix internals — the same capabilities as the built-in AI Assistant.
The bridge only listens on
127.0.0.1(localhost) — no network exposureAuth is handled via a bearer token auto-generated each time Selenix starts
Connection config is stored at
~/.selenix/bridge.jsonand re-read on every call, so restarting Selenix does not require restarting Claude Desktop
Available Tools
Reading & Inspection
Tool | Description |
| Capture a screenshot of the browser window |
| Get the page HTML with interactive elements annotated |
| Search page HTML by regex pattern |
| Get DOM context for a command's target element |
| Get all commands in the current test |
| List all tests in the project |
| Get a test by its name |
| List all available Selenix commands |
| Get detailed docs for a specific command |
| Get project metadata |
| Get the active test suite |
| Get pass/fail results from the last test run |
| Get recent logs (all types), 20 per page, most recent first. Optional |
| Get a summary of the current workspace |
Writing & Execution
Tool | Description |
| Add commands to a test at a specific position |
| Run a test and wait for results (up to 2 minutes) |
| Replace all commands in a test |
| Apply targeted fixes — update, remove, or insert commands |
Troubleshooting
"Cannot read Selenix bridge config" Selenix isn't running or the MCP Server isn't enabled. Open Selenix → System Settings → set Enable MCP Server to Yes.
"Cannot connect to Selenix bridge" Selenix may have just restarted. Try the request again — the MCP server will automatically pick up the new connection details.
Tools not appearing in Claude Desktop Make sure you restarted Claude Desktop after editing the config file. Verify the config JSON is valid (no trailing commas, correct brackets).
Claude Desktop won't open after config change The config JSON is likely malformed. Open the config file in a text editor, fix the JSON syntax, and try again. Common issues: missing comma between entries, trailing comma after the last entry.
Uninstall
npm uninstall -g @selenix/mcp-serverThen remove the "selenix" entry from your Claude Desktop config file and restart Claude Desktop.
Available Tools
18 toolsadd_commandsC
Add one or more commands to a test at a specific index. Commands use camelCase names (e.g., "click", "type", "open", "executeScript").
Inputs: test_id (required), index (required), commands (required).
clear_and_replace_commandsA
Replace ALL commands in a test with a new set. This removes all existing commands first.
Inputs: test_id (required), commands (required).
fix_commandsB
Apply targeted fixes to commands in a test: update, remove, or insert at specific indexes. Indexes refer to the current command list BEFORE any fixes are applied.
Inputs: test_id (required), fixes (required).
get_active_suiteB
Get information about the currently active test suite and its tests.
Inputs: no named parameters.
get_command_infoB
Get detailed documentation for a specific Selenix command, including target and value parameter descriptions.
Inputs: command_name (required).
get_command_listB
List all available Selenix commands (click, type, open, etc.) with descriptions.
Inputs: no named parameters.
get_current_testB
Get the commands of the current test (or a specific test by ID). Returns test name, ID, and all commands with their targets and values.
Inputs: test_id.
get_logsA
Get recent logs from Selenix (all types). Returns 20 logs per page, most recent first. Use page parameter to paginate (0 = most recent, 1 = next 20, etc.).
Inputs: page.
get_page_htmlA
Get the HTML of the web page in the active playback window, with interactive elements annotated. Scripts and styles are removed.
Inputs: tab_id.
get_project_infoB
Get metadata about the current Selenix project: name, base URL, test count, suite count.
Inputs: no named parameters.
get_screenshotA
Capture a screenshot of the active browser window in Selenix playback. Returns a base64-encoded JPEG image.
Inputs: no named parameters.
get_search_htmlC
Search page HTML by regex pattern and return surrounding HTML context for matches.
Inputs: regex (required), index, tab_id.
get_selected_command_infoA
Get comprehensive context about a command target element: DOM hierarchy, accessibility info, surrounding HTML, interactive properties.
Inputs: command_id.
get_step_resultsC
Get pass/fail/error results for each command in the last test run.
Inputs: test_id.
get_test_by_nameB
Get a specific test by its name, including all commands.
Inputs: test_name (required).
get_test_listB
List all tests in the current Selenix project with their IDs, names, and command counts.
Inputs: no named parameters.
get_workspace_contextA
Get a summary of the current Selenix workspace: project name, active test, test count, selected command count.
Inputs: no named parameters.
run_testA
Run a test and wait for it to complete. Returns the pass/fail result for each command. This operation may take up to 2 minutes.
Inputs: test_id.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 18 tools
Most tools have clearly distinct purposes, such as 'add_commands' for adding commands versus 'clear_and_replace_commands' for replacing all commands. However, some overlap exists between 'get_current_test' and 'get_test_by_name', as both retrieve test details, which could cause minor confusion. Overall, the descriptions help clarify boundaries, but the duplication in test retrieval tools slightly reduces clarity.
Tool names follow a highly consistent verb_noun pattern throughout, such as 'add_commands', 'get_active_suite', and 'run_test'. All tools use snake_case without deviation, making them predictable and easy to parse. This consistency enhances usability and reduces cognitive load for agents.
With 18 tools, the count is slightly high but reasonable for a Selenium automation domain, covering test management, command operations, and debugging. It provides comprehensive functionality without feeling overly bloated. A minor reduction could improve focus, but the scope justifies most tools.
The tool set offers complete coverage for Selenium test automation, including CRUD operations for commands (add, clear, fix), test retrieval and execution (get, run), and debugging features (logs, screenshots, HTML analysis). No obvious gaps exist; agents can manage the full test lifecycle from creation to analysis seamlessly.
Maintenance
Related MCP Connectors
Official Browserless MCP server: browser automation, scraping, screenshots and PDFs for AI agents.
Hosted Web Crawler MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
MCP server for Mint — AI-powered QA that runs your app in a real browser on every PR.
Hosted Magento MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Related MCP Servers
- AlicenseCqualityDmaintenanceAn MCP server that gives Claude Code real browser control for web automation, testing, and screenshots.3246 npm158MIT
- AlicenseAqualityCmaintenanceAn MCP server that gives Claude direct control of a real browser via Playwright, enabling AI-driven web testing, autonomous test execution, and live failure analysis through natural language.83MIT
- AlicenseNot gradedqualityDmaintenanceLocal browser automation MCP server for Claude Code, enabling navigation, clicking, typing, and inspecting real web pages via Google Chrome.1MIT
- AlicenseNot gradedqualityAmaintenanceAn MCP server that gives Claude Code a real browser to navigate, click, type, screenshot, and inspect pages via accessibility snapshots. It also runs multi-step test scenarios with assertions, session recording, visual regression, and console/network monitoring from the terminal.106 npmMIT