OpenChrome MCP Server: Control Your Real Chrome with AI

• Opens a web address or goes back and forward

  Loads a URL in the browser tab, or moves forward or back in history.

  Try: “Open https://example.com”

  Write · navigate

• Moves the mouse, types and takes screenshots

  Does mouse and keyboard actions on a tab, including clicks, typing and screenshots.

  Try: “Click the Sign in button and take a screenshot”

  Write · computer

• Reads a page as text, DOM or accessibility tree

  Returns the page as DOM, an accessibility tree, a CSS diagnostic, a short semantic summary or clean markdown.

  Try: “Read this page as markdown”

  Read · read_page

• Finds elements on the page

  Looks for elements that match a query and returns them so the assistant can act on them.

  Try: “Find the checkout button”

  Read · find

• Looks up elements by CSS selector or XPath

  Runs a CSS selector or XPath query on the page and returns the matching elements.

  Try: “Get all links inside the nav bar”

  Read · query_dom

• Clicks or hovers on an element you describe

  Finds an element from a plain-language description, clicks, hovers or double-clicks it, waits for the page to settle and returns the new state.

  Try: “Click the "Add to cart" button”

  Write · interact

• Sets the value of a form field

  Sets one form element to a value using a reference from the page.

  Try: “Set the email field to test@example.com”

  Write · form_input

• Lists the tabs OpenChrome manages

  Returns the tab IDs of the current session, grouped by worker.

  Try: “What tabs are open?”

  Read · tabs_context

• Checks a tab or pauses input for you

  Observes a managed tab, checks conditions you supply, or pauses automatic input so you can take control.

  Try: “Pause automation so I can log in myself”

  Write · oc_browser_control

• Opens a new tab

  Creates a new tab at a URL.

  Try: “Open a new tab at example.com”

  Write · tabs_create

• Closes tabs

  Closes one or more tabs by tab ID or worker ID.

  Try: “Close the tabs you opened”

  Delete · tabs_close

• Waits for a condition

  Waits until a condition on the page is true before the next step.

  Try: “Wait until the results table appears”

  Read · wait_for

• Takes a screenshot of the page

  Saves a screenshot to a file or returns it as base64.

  Try: “Screenshot this page”

  Write · page_screenshot

• Shows the Chrome connection status

  Returns connection health, such as reconnect count, ping time and state.

  Try: “Is the Chrome connection healthy?”

  Read · oc_connection_health

• Shuts OpenChrome and Chrome down

  Stops OpenChrome and closes the Chrome it launched.

  Try: “Stop OpenChrome and close Chrome”

  Delete · oc_stop

• Runs JavaScript in the page

  Executes JavaScript you supply in the context of the current page.

  Try: “Run document.title in the page”

  Runs code · javascript_tool

• Reads, sets and deletes cookies

  Gets, sets, deletes or clears the browser's cookies.

  Try: “Show cookies for this site”

  Delete · cookies

• Reads and changes localStorage and sessionStorage

  Manages the page's local and session storage.

  Try: “Show localStorage for this page”

  Delete · storage

• Returns the HTML of a page or element

  Gets the HTML content of the page or one element.

  Try: “Get the HTML of the main section”

  Read · page_content

• Saves the page as a PDF

  Generates a PDF of the page.

  Try: “Save this page as a PDF”

  Write · page_pdf

• Records requests with response bodies

  Captures network requests including response bodies, up to a size cap.

  Try: “Capture the API responses while I load the page”

  Read · network_capture_full

• Logs, blocks or edits network requests

  Intercepts requests so you can log, block or modify them.

  Try: “Block all image requests on this page”

  Write · request_intercept

• Sets HTTP login credentials

  Sets or clears HTTP basic-auth credentials for the browser.

  Try: “Set HTTP auth for the staging site”

  Write · http_auth

• Uploads a file into a page

  Attaches a file to a file input element on the page.

  Try: “Upload ./report.pdf to the file field”

  Write · file_upload

• Fills several fields and can submit

  Fills form fields and can submit the form.

  Try: “Fill the signup form and submit it”

  Write · fill_form

• Does several browser steps from one instruction

  Runs a multi-step browser action from a plain-language instruction.

  Try: “Search for "laptops" and open the first result”

  Write · act

• Crawls a website

  Recursively crawls a site breadth first.

  Try: “Crawl the docs section”

  Network · crawl

• Pulls structured data from a page

  Extracts data matching a JSON schema from JSON-LD, Microdata, OpenGraph or CSS.

  Try: “Extract the product schema from this page”

  Read · extract_data

• Generates a 2FA code

  Generates the current one-time code (TOTP) for a domain.

  Try: “Get the 2FA code for example.com”

  Read · oc_totp_generate

• Exports the tab's login state

  Exports cookies, local and session storage and optional user agent and viewport as a portable plain-text envelope.

  Try: “Export my login state for this tab”

  Read · oc_context_export

• Imports saved login state into a tab

  Replaces the tab's auth state with an envelope made by the export tool.

  Try: “Import the saved context into this tab”

  Write · oc_context_import

• Searches the tool call history

  Queries the journal of tool calls made in the session.

  Try: “Show my last 10 tool calls”

  Read · oc_journal

• Detects CAPTCHA and login gates

  Checks whether the current tab is blocked by a CAPTCHA, bot check, SSO redirect, paywall or 2FA prompt.

  Try: “Is this page behind a login wall?”

  Read · oc_gate_inspect

Watch out for

• It can use every account your Chrome is logged in to

  The security policy says the assistant gets access to the browser's authenticated sessions, such as banking and email. Use a dedicated Chrome profile and keep sensitive accounts out of it. github.com

• It can read and export cookies and login state

  The cookies, storage and context export tools hand login tokens to the assistant, and a TOTP tool generates 2FA codes. These load on demand, so skip or disable them if you do not need them. github.com

• It runs JavaScript in pages

  javascript_tool executes code you or the assistant supply in the page, and request_intercept can change traffic. There is no read-only mode. github.com

• Web pages can try to steer the assistant

  The project states it has no way to detect prompt injection from visited pages. Use the domain blocklist and audit log options, and review what the assistant does. github.com

• It writes files and keeps local state

  Screenshots, PDFs, evidence bundles, recordings and a task ledger are saved on your machine. github.com

Security advisories