Firefox DevTools MCP Server: Browse and Debug Firefox with AI

• Lists the open Firefox tabs

  Returns each open tab with its number, title and address, and marks the one the assistant is working in.

  Try: “What tabs are open in Firefox right now?”

  Read · list_pages

• Opens a new tab at an address

  Opens a tab at the URL you give and returns its tab number. You can choose whether it waits for nothing, for the page structure, or for the full load.

  Try: “Open https://example.com in a new tab”

  Write · new_page

• Sends the current tab to an address

  Loads a new URL in the tab the assistant is working in, with the same choice of how long to wait.

  Try: “Go to https://example.com/pricing”

  Write · navigate_page

• Switches to another tab

  Picks the active tab by number, or by part of its address or title. The number wins if you give both.

  Try: “Switch to the tab with GitHub in the title”

  Write · select_page

• Closes a tab

  Closes the tab with the number you give.

  Try: “Close tab 2”

  Delete · close_page

• Reads the visible text of a page

  Returns the page text as the browser shows it, cut off at 20,000 characters unless you ask for a file. It can save the full text to a file.

  Try: “Get the text of this page and save it to a file”

  Write · get_page_text

• Captures a page outline with element IDs

  Builds a short outline of the page where each element gets an ID that stays valid until the element goes away. Click and fill tools use those IDs. It can write the full outline to a file.

  Try: “Take a snapshot of the checkout form”

  Write · take_snapshot

• Turns an element ID into a CSS selector

  Looks up an ID from a snapshot and returns the CSS selector for that element. It fails if the element is gone.

  Try: “Give me a CSS selector for the Submit button”

  Read · resolve_uid_to_selector

• Forgets the saved element IDs

  Clears the element IDs from earlier snapshots. The README says it is usually not needed.

  Try: “Clear the snapshot IDs”

  Write · clear_snapshot

• Clicks an element

  Clicks the element with the given ID. Turn on double-click to click twice.

  Try: “Click the Sign in button”

  Write · click_by_uid

• Moves the pointer over an element

  Hovers over the element with the given ID, for menus and tooltips.

  Try: “Hover over the Products menu”

  Write · hover_by_uid

• Fills in a text field

  Puts text into the input or text area with the given ID.

  Try: “Fill the email field with test@example.com”

  Write · fill_by_uid

• Drags one element onto another

  Sends HTML5 drag events from one element to another, for drag-and-drop interfaces.

  Try: “Drag the first card onto the Done column”

  Write · drag_by_uid_to_uid

• Fills several fields at once

  Takes a list of element ID and value pairs and fills them all in one call.

  Try: “Fill the name, email and phone fields in the signup form”

  Write · fill_form_by_uid

• Attaches a local file to a file input

  Sets a file input to a path on your computer, so the page can upload it when the form is sent.

  Try: “Upload ./resume.pdf to the file input”

  Write · upload_file_by_uid

• Presses a key or shortcut

  Presses one key, with optional modifiers such as ctrl+shift+t, to submit, dismiss or trigger a shortcut. Typing text goes through the fill tool instead.

  Try: “Press Escape to close the dialog”

  Write · press_key

• Types text key by key

  Types into the focused element one key at a time and can press a key such as Enter afterwards. It suits autocomplete fields and rich text editors that ignore a plain fill.

  Try: “Type "laptop" in the search box and press Enter”

  Write · type_text

• Lists the requests the page made

  Lists captured network requests with filters for address, method, status, type and time, and can save every match to a file as JSON. Each result gets an ID for the detail tool.

  Try: “Show failed API requests from the last minute”

  Write · list_network_requests

• Shows one request in detail

  Returns headers and the response body for one request, and the request body when there is one. Long text bodies are cut and binary bodies are summarized. It can save the full record to a file.

  Try: “Show the response body for request 12”

  Write · get_network_request

• Turns the browser cache off or on

  Switches between normal caching and bypassing the cache, for the current tab or the whole browser, so a changed file is not hidden by an old copy.

  Try: “Bypass the cache and reload the page”

  Write · set_network_cache

• Lists console messages

  Returns browser console messages filtered by level, time, text or source, and can save all matches to a file.

  Try: “Show console errors since the page loaded”

  Write · list_console_messages

• Empties the console log

  Clears the console messages the server has collected.

  Try: “Clear the console messages”

  Write · clear_console_messages

• Takes a screenshot of the page

  Captures the visible area, or the whole scrollable page, as a PNG. It can save the image to a file instead of returning it.

  Try: “Take a full-page screenshot”

  Write · screenshot_page

• Takes a screenshot of one element

  Captures a single element, by ID from a snapshot, as a PNG. It can save the image to a file.

  Try: “Screenshot just the pricing table”

  Write · screenshot_by_uid

• Saves the page as a PDF

  Prints the current page to a PDF using its print styles and returns the file path. Files go under ~/.firefox-devtools-mcp/output by default.

  Try: “Print this page to PDF in landscape”

  Write · print_page

• Lists downloads since startup

  Shows downloads the server has seen, with status and saved file path, and can filter by status or address.

  Try: “What has been downloaded so far?”

  Read · list_downloads

• Clears the download list

  Empties the server's record of tracked downloads.

  Try: “Clear the downloads list”

  Write · clear_downloads

• Allows, blocks or resets downloads

  Lets downloads save to a folder you name, cancels them, or goes back to the browser default, without the save dialog. The folder is created if missing.

  Try: “Allow downloads into ./downloads”

  Write · set_download_behavior

• Accepts a browser dialog

  Clicks OK on an alert, confirm or prompt dialog, and can enter text for a prompt.

  Try: “Accept the confirm dialog”

  Write · accept_dialog

• Dismisses a browser dialog

  Closes a dialog with the cancel action.

  Try: “Dismiss the dialog”

  Write · dismiss_dialog

• Goes back or forward

  Moves the tab one step back or forward in history. Saved element IDs go stale after this.

  Try: “Go back one page”

  Write · navigate_history

• Resizes the page area

  Sets the viewport width and height in pixels for the current tab, to test responsive layouts.

  Try: “Set the viewport to 390 by 844”

  Write · set_viewport_size

• Reads Firefox's own log output

  Returns recent stdout and stderr from the Firefox process, including crash and warning lines, with line count, text filter and time window options.

  Try: “Show the last 50 Firefox log lines containing "error"”

  Files · get_firefox_output

• Shows how Firefox was started

  Returns the Firefox binary path, environment variables and the output file location for the running instance.

  Try: “Which Firefox binary is the server using?”

  Read · get_firefox_info

• Ends the browser session

  Closes Firefox if the server started it. If the server attached to your existing Firefox, it just lets go and leaves the browser running.

  Try: “Close the Firefox session”

  Delete · close_firefox_session

• Installs a Firefox extension

  Installs an extension from a .xpi or .zip file, base64 data or an unpacked folder using WebDriver BiDi. The install is temporary unless you ask for a permanent one, which needs a signed extension.

  Try: “Install the extension from ./my-addon as a temporary add-on”

  Write · install_extension

• Removes a Firefox extension

  Uninstalls an extension by its ID, which comes from the install tool.

  Try: “Uninstall addon@example.com”

  Delete · uninstall_extension

• Checks whether the profiler is recording

  Reports whether the Firefox profiler is running right now.

  Try: “Is the profiler recording?”

  Read · profiler_is_active

• Starts the Firefox profiler

  Starts recording a performance profile, from a named preset such as web-developer or networking or from entries, interval, features and threads you give it.

  Try: “Start the web-developer profiler preset”

  Write · profiler_start

• Stops the profiler and saves the profile

  Ends the recording and saves the profile file in the downloads folder, then returns its path. A discard option throws the recording away.

  Try: “Stop the profiler and save the profile”

  Write · profiler_stop

• Starts recording a video of the page

  Records the page viewport to a video file (WebM by default) in the downloads folder and returns an ID for stopping it. It needs Firefox 154 or newer.

  Try: “Record a screencast of this checkout flow”

  Write · screencast_start

• Stops a screen recording

  Finishes a running recording and returns the path of the video file.

  Try: “Stop the screencast”

  Write · screencast_stop

• Runs JavaScript in the page

  Runs a JavaScript function you supply in the page and returns the result. It can run on any site you opened and can save its output to a file.

  Try: “Run a function that returns document.title”

  Runs code · evaluate_script

• Turns on the JavaScript debugger

  Switches on the debugger for the current page. It must run before logpoints work and needs Firefox 153 or newer.

  Try: “Enable the debugger for this page”

  Write · enable_debugger

• Lists the page's JavaScript files

  Returns every script file loaded in the page. The debugger must be enabled first.

  Try: “List the scripts this page loaded”

  Read · list_scripts

• Reads a script file's source

  Fetches the source of a loaded JavaScript file by address. The debugger must be enabled first.

  Try: “Show the source of app.js”

  Read · get_script_source

• Logs an expression at a code line

  Adds a logpoint at a script URL and line. Each time that line runs, the expression is evaluated and stored without pausing the page.

  Try: “Set a logpoint on line 42 of app.js that logs the cart total”

  Write · set_logpoint

• Removes a logpoint

  Deletes a logpoint you set earlier, by its ID.

  Try: “Remove logpoint 1”

  Delete · remove_logpoint

• Reads what a logpoint collected

  Returns the values a logpoint stored since it was set.

  Try: “What did logpoint 1 record?”

  Read · get_logpoint_results

Watch out for

• It can run any JavaScript in the pages you open

  The default tool preset includes evaluate_script, so a page that manipulates the assistant could read page data or act as you on that site. Start the server with the slim preset to remove it; slim still has click, fill, file upload and tab-closing tools. github.com

• Pointed at your normal profile, it can use your logins

  The connect-existing option attaches to a running Firefox with your cookies and sessions. Run the server with its own Firefox profile instead. github.com

• It can install and remove Firefox extensions

  install_extension and uninstall_extension are part of the default preset. Leave out the management modules with an explicit tools list if you do not want this. github.com

• It writes files to your disk

  Screenshots, PDFs, profiles, recordings and the saveTo option write under ~/.firefox-devtools-mcp/output by default. The unrestricted-save-paths flag lets the assistant write anywhere, so leave it off. github.com

• Android mode wipes the app's data

  The Android option runs a data clear on the target Firefox app before each session, which removes tabs, history, bookmarks and passwords. Use a Firefox build set aside for automation. github.com

Security advisories