Atlas / MCP servers / shiplightai / shiplightai

shiplightaiBLOCK

Name: shiplightai
Author: shiplightai

mcp/shiplightai/shiplight-cli

MCP server by shiplightai · Audited 2026-10-12

AI-powered end-to-end testing for Playwright. Write tests as YAML with natural-language steps, with self-healing locators and a visual debugger.

Verdict

BLOCK

Grade

Trust score

47 /100

Exposed tools

74 57r · 12w · 5d

Transport

stdio

License

MIT

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

AI-powered end-to-end testing for Playwright. Write tests as YAML with natural-language steps, run them alongside your existing .test.ts files, and get self-healing locators and a visual debugger.

Documentation: [docs.shiplight.ai](https://docs.shiplight.ai)

What ships from this repository

Quick start

npx shiplightai@latest create ./my-tests
cd my-tests
npm install
npx playwright install chromium
npx shiplight test

A test looks like this:

goal: A shopper can add an item to the cart
base_url: https://example.com

statements:
- URL: /products

# Natural language: resolved by a model on the first run
- intent: Click the "Add to cart" button on the first product

# Once resolved, a step carries its locator and replays deterministically
- intent: Open the cart
action: click
locator: "getByRole('link', { name: 'Cart' })"

- VERIFY: The cart shows exactly one item

A statement written as plain intent is resolved by a model at runtime; once resolved it carries a concrete locator, so later runs replay deterministically and only fall back to the model when the page has changed under it.

Self-hosting

Run the CLI and browsers on your own machines or CI runners, with your own AI provider key and the action cache and reports in storage you control: apps/cli/docs/self-hosting.md.

Shiplight Cloud shuts down on October 31, 2026. After that date this is the only way to run Shiplight, and SHIPLIGHT_API_TOKEN stops working.

Working on the code

Requires Node.js >= 22 and p

Read from source at commit 32099b12f2caOBSERVED · 2026-10-12

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code (npm)

claude mcp add mcp -- npx -y @shiplightai/[email protected]

Exposed tools (74)

57 read · 12 write · 5 destructive. Blast radius: 5 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`act`	write	Execute browser actions using element indices from inspect_page. CRITICAL: Before your first act call, you MUST read resource
`ai_action`	read	Use AI to perform a specific action on the page. MCP-only tool.
`ai_assert`	read	Assert that a statement is true based on the current screen state
`ai_extract`	write	Use AI to extract information from the page and save it to a variable. MCP-only tool.
`ai_step`	write	Use AI to execute a complete multi-action step. MCP-only tool.
`ai_wait_until`	read	Wait until a specific condition is met on the page (e.g.,
`attach_to_browser`	read	Connect to a browser for testing. Two modes:\n
`back`	read	Press the back button
`browser-session-basics`	read	Basic browser session management for UI verification
`clear_execution_history`	destructive	Clear the execution history for a session. The execution history tracks past browser operations within a session. Clear it to start fresh or reduce memory usage.
`clear_input`	destructive	Clear text from an input field
`clear_logs`	destructive	Clear console and network logs for a session
`click`	read	Click an interactive element.
`close_all`	read	Close all browser sessions
`close_session`	read	Close a browser session. Omit session_id entirely to close ALL open sessions.
`close_tab`	read	Close a browser tab by index. If no index is specified, closes the current tab.
`constrained`	read	field with a refinement
`done`	read	Mark the task as complete
`double_click`	read	Double-click an interactive element.
`double_click_at`	read	Double-click at the given normalized coordinate.
`double_tap`	read	Double-tap an element
`enter`	read	Press the enter key
`extract_email_code`	read	Extract verification code from email inbox
`extract_email_content`	read	Extract verification codes, magic links, or custom content from emails received via Mailgun. Polls the Mailgun inbox for matching emails and uses AI to extract the requested content.
`generate_2fa_code`	read	Generate a 2FA code from an OTP secret key. The code is saved to the
`generate_html_report`	read	Generate a self-contained HTML session report.
`get_browser_console_logs`	read	Get console logs from the browser (errors, warnings, etc.)
`get_browser_network_logs`	read	Get network request logs from the browser
`get_dropdown_options`	read	Get all options from a native dropdown
`get_local_artifact`	read	Get a local artifact file (screenshot, DOM snapshot, etc.) from a live browser session. Only accepts local file paths, NOT S3 URIs. For test run artifacts, use get_step_artifacts instead.
`get_page_info`	read	Get basic page information (URL and title) without taking a screenshot. Use this as a lightweight way to check what page you
`get_relay_status`	read	Get extension relay server status and tab mapping
`get_session_state`	read	Get current state of a session (URL, session type)
`go_back`	read	Navigate back to the previous page in browser history.
`go_to_url`	write	Navigate to a specific URL. Use this to visit web pages. Set new_tab=true to open in a new tab.
`home`	read	Press the home button
`hover`	read	Hover over an interactive element.
`input`	read	Type text into the currently focused input field
`input_text`	read	Type text into an input field
`js_code`	write	Execute custom JavaScript code in the browser context. Use for advanced interactions not covered by other tools.
`long_press`	read	Long press an element
`navigate`	read	Navigate to a URL in an existing browser session. Examples: - Navigate to a specific page: url=
`new_session`	write	Create a new browser session with optional device emulation. Returns a session_id for subsequent operations.
`opt_out`	read	no resolveArgs
`perform_accurate_operation`	read	Perform an operation that requires accurate interaction like dragging or interacting with a specific area of an element. Only use this action when neccecary.
`pick_role`	read	enum message embeds the received value
`plain_enum`	read	no redactor
`press`	read	Press a single keyboard key or key combination. Supports combinations like
`press_key`	read	Press a system key by key code
`raw_args`	read	reports rawArgs
`reload_page`	destructive	Reload the current page. WARNING: This will reset any in-progress forms or multi-step flows (like login). Only use when you need to refresh stale data or clear a stuck state. Do NOT use during login, checkout, or form submission flows.
`right_click`	read	Right-click an interactive element.
`right_click_at`	read	Right-click at the given normalized coordinate.
`save_storage_state`	write	Save the browser session
`save_variable`	write	Save a value to a named variable for later use in the test.
`scroll`	read	Scroll content with precise from/to coordinates (preferred for scrolling lists, feeds, etc.)
`scroll_on_element`	read	Scroll on a scrollable element horizontally and vertically.
`scroll_to_element`	read	Scroll to find an element
`scroll_to_text`	read	Scroll the page until specified text becomes visible.
`select_dropdown_option`	read	Select an option from a dropdown/select element by value, label, or index.
`set_date_for_native_date_picker`	write	Set a date in a native HTML date picker input (<input type=
`swipe`	read	Swipe gesture for switching tabs, dismissing dialogs, or navigating pages (use scroll for scrolling content)
`swipe_points`	destructive	Drag/swipe between two specific points (for drag-and-drop, picture-in-picture, etc.)
`switch_tab`	read	Switch to a different browser tab by index (0-based).
`tap`	read	Tap an element
`unresolvable`	read	placeholder has no variable
`update_variables`	write	Update session variables for use in subsequent browser operations. Variables are key-value pairs that can be referenced in browser tasks. Mark sensitive values (passwords, tokens) with sensitive_keys to prevent logging. Example: variables: { username:
`upload_file`	write	Click on target element (usually contains hints like
`upload_file_to_file_input`	write	Upload a file to a file input element. The file path is relative to the test project root.
`verify`	read	Verify that a statement is true based on the current page state. Use this to check assertions about the page content, element visibility, or any condition.
`wait`	read	Wait for a specified duration
`wait_for_download_complete`	read	Wait for an in-progress download to complete.
`wait_for_page_ready`	read	Wait for the page to be fully loaded (network idle and DOM ready). Use this when the page is still loading or not ready for interaction.
`with_description`	read	description is stripped before resolution

Trust audit

BLOCKgrade F · trust 47/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	PASS
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	PASS
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED