Firefox DevTools MCP
Allows interaction with a Firefox browser via WebDriver BiDi, providing tools for navigation, page snapshots, input handling, network monitoring, console inspection, screenshots, and browser management.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Firefox DevTools MCP navigate to mozilla.org and list console errors"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Firefox DevTools MCP
Model Context Protocol server for automating Firefox via WebDriver BiDi (through Selenium WebDriver). Works with Claude Code, Claude Desktop, Cursor, Cline and other MCP clients.
Repository: https://github.com/mozilla/firefox-devtools-mcp
Note: This MCP server requires a local Firefox browser installation and cannot run on cloud hosting services like glama.ai. Use
npx @mozilla/firefox-devtools-mcp@latestto run locally, or use Docker with the provided Dockerfile.
Security
Browser MCP servers carry inherent risks. A few key practices:
Use a dedicated Firefox profile. Never run the server against your regular profile — the agent has access to whatever the browser can reach, including cookies and saved sessions.
Be cautious about which sites you visit. Pages can return content designed to manipulate the agent (prompt injection). Stick to sites you control or trust.
Enable only the tool modules you need. Higher presets such as
--tool-preset developer(script, debugging) and--tool-preset mozilla(privileged context) significantly expand what the agent can do.
See SECURITY.md for a full breakdown of risks and how to report vulnerabilities.
Related MCP server: MCP Selenium Server
Requirements
Node.js ≥ 20.19.0
Firefox 100+ installed (auto‑detected, or pass
--firefox-path)
Install and use with Claude Code (npx)
Recommended: use npx so you always run the latest published version from npm.
Option A — Claude Code CLI
claude mcp add firefox-devtools npx @mozilla/firefox-devtools-mcp@latestPass options either as args or env vars. Examples:
# Headless + viewport via args
claude mcp add firefox-devtools npx @mozilla/firefox-devtools-mcp@latest -- --headless --viewport 1280x720
# Or via environment variables
claude mcp add firefox-devtools npx @mozilla/firefox-devtools-mcp@latest \
--env START_URL=https://example.com \
--env FIREFOX_HEADLESS=trueOption B — Edit Claude Code settings JSON
Add to your Claude Code config file:
macOS:
~/Library/Application Support/Claude/Code/mcp_settings.jsonLinux:
~/.config/claude/code/mcp_settings.jsonWindows:
%APPDATA%\Claude\Code\mcp_settings.json
{
"mcpServers": {
"firefox-devtools": {
"command": "npx",
"args": ["-y", "@mozilla/firefox-devtools-mcp@latest", "--headless", "--viewport", "1280x720"],
"env": {
"START_URL": "about:blank"
}
}
}
}Option C — Helper script (local dev build)
npm run setup
# Choose Claude Code; the script saves JSON to the right pathTry it with MCP Inspector
npx @modelcontextprotocol/inspector npx @mozilla/firefox-devtools-mcp@latest --start-url https://example.com --headlessThen call tools like:
list_pages,select_page,navigate_pagetake_snapshotthenclick_by_uid/fill_by_uidlist_network_requests(always‑on capture),get_network_requestlist_downloads(always‑on capture),set_download_behaviorscreenshot_page,list_console_messages
CLI options
You can pass flags or environment variables (names on the right):
--firefox-path— absolute path to Firefox binary--headless— run without UI (FIREFOX_HEADLESS=true)--viewport 1280x720— initial window size--profile-path— use a specific Firefox profile--firefox-arg— extra Firefox arguments (repeatable)--start-url— open this URL on start (START_URL)--accept-insecure-certs— ignore TLS errors (ACCEPT_INSECURE_CERTS=true)--connect-existing— attach to an already-running Firefox instead of launching a new one (CONNECT_EXISTING=true)--marionette-port— Marionette port for connect-existing mode, default 2828 (MARIONETTE_PORT)--pref name=value— set Firefox preference at startup viamoz:firefoxOptions(repeatable)--tool-preset— select which tool modules to enable:slim,basic(default),developer,mozilla, orall. See Tool modules and presets. (TOOL_PRESET)--tools— explicit list of tool modules to enable, overriding--tool-presetentirely (e.g.--tools pages network script). See Tool modules and presets.--enable-script— deprecated, use--tool-preset developeror--tools ... script debugging. Selects thedevelopertool preset. (ENABLE_SCRIPT=true)--enable-privileged-context— deprecated, use--tool-preset mozillaor--tools ... privileged prefs. Selects themozillatool preset. RequiresMOZ_REMOTE_ALLOW_SYSTEM_ACCESS=1(ENABLE_PRIVILEGED_CONTEXT=true)--android-device— enable Firefox for Android mode; value is the ADB device serial (e.g.emulator-5554). Runadb devicesto list connected devices. Omit the value or useautoto select the single connected device automatically.--android-package— Android app package name, defaultorg.mozilla.firefox. Other packages:org.mozilla.firefox_betafor Firefox Beta,org.mozilla.fenixfor Firefox Nightly,org.mozilla.fenix.debugfor Firefox Nightly Debug,org.mozilla.geckoview_examplefor geckoview (ANDROID_PACKAGE)--log-file— write MCP server logs to a file instead of stderr. Useful for debugging sessions with MCP clients that hide server output. SetDEBUG=*to also include verbose debug logs. Example:--log-file /tmp/firefox-mcp.log
Tool modules and presets
Tools are grouped into modules. You choose which modules to expose either with a named preset
(--tool-preset) or with an explicit list (--tools). When both are given, --tools wins and
the preset is ignored.
Modules: pages, snapshot, input, network, console, screenshot, utilities,
management, webextension, profiler, screencast, script, debugging, prefs,
privileged.
Presets (each is a superset of the previous):
slim—pages,snapshot,input,network,consolebasic(default) —slimplusscreenshot,utilities,management,webextension,profiler,screencastdeveloper—basicplusscript,debuggingmozilla—developerplusprefs,privilegedall— every module
# Use the developer preset (adds script and debugging tools)
npx @mozilla/firefox-devtools-mcp --tool-preset developer
# Enable only the modules you need
npx @mozilla/firefox-devtools-mcp --tools pages network consoleThe prefs and privileged modules require MOZ_REMOTE_ALLOW_SYSTEM_ACCESS=1 and are only
available in the Mozilla-internal build; the public package silently skips them even if requested.
Useful preferences (--pref)
remote.prefs.recommended=false. When Firefox runs in automation, it applies RecommendedPreferences that modify browser behavior for testing. Set remote.prefs.recommended to false to skip those and have a configuration closer to a regular Firefox instance.
remote.log.level=Trace. Enable verbose WebDriver protocol logs in Firefox. The MCP server will automatically pass the matching log level to geckodriver so both sides log at the same verbosity.
app.update.disabledForTesting=false. Allow Firefox to automatically download and apply updates. Note that updates may interrupt your session. Requires also setting remote.prefs.recommended=false.
Firefox for Android
Use --android-device to automate Firefox running on an Android device. Requires adb on your PATH and geckodriver, which is managed automatically.
# List connected devices
adb devices
# Launch Firefox for Android on the single connected device
npx @mozilla/firefox-devtools-mcp --android-device auto
# Target a specific device
npx @mozilla/firefox-devtools-mcp --android-device <serial>
# Use Firefox Nightly instead
npx @mozilla/firefox-devtools-mcp --android-device <serial> --android-package org.mozilla.fenixPort forwarding between the host and device is handled automatically by geckodriver.
Connect to existing Firefox
Use --connect-existing to automate your real browsing session, with cookies, logins, and open tabs intact:
# Start Firefox with Marionette and the Remote Agent (BiDi)
firefox --marionette --remote-debugging-port
# Run the MCP server
npx @mozilla/firefox-devtools-mcp --connect-existing --marionette-port 2828Both flags are required because the MCP uses both WebDriver Classic (--marionette) and WebDriver BiDi (--remote-debugging-port). If Firefox is only started with --marionette, the MCP server fails to connect and asks you to restart Firefox with both flags.
Warning: Do not leave Marionette enabled during normal browsing. It sets
navigator.webdriver = trueand changes other browser fingerprint signals, which can trigger bot detection on sites protected by Cloudflare, Akamai, etc. Only enable Marionette when you need MCP automation, then restart Firefox normally afterward.
Tool overview
Pages: list/new/navigate/select/close
Snapshot/UID: take/resolve/clear (take supports optional
saveTo)Input: click/hover/fill/drag/upload/form fill <<<<<<< HEAD
Network: list/get (ID‑first, filters, always‑on capture; both support optional
saveTo)Console: list/clear (list supports optional
saveTo) =======Network: list/get (ID‑first, filters, always‑on capture)
Downloads: list_downloads/clear_downloads (always‑on capture), set_download_behavior (allow/deny/default)
Console: list/clear
6873767 (Bug 2055949 - [firefox-devtools-mcp] expose download events to firefox devtools mcp)
Screenshot: page/by uid (with optional
saveTofor CLI environments)Script: evaluate_script (with optional
saveTofor bulky results)Privileged Context: list/select privileged ("chrome") contexts, evaluate_privileged_script (requires
MOZ_REMOTE_ALLOW_SYSTEM_ACCESS=1)WebExtension: install_extension, uninstall_extension, list_extensions (list requires
MOZ_REMOTE_ALLOW_SYSTEM_ACCESS=1)Firefox Management: get_firefox_info, get_firefox_output, restart_firefox, set_firefox_prefs, get_firefox_prefs
Profiler: profiler_is_active, profiler_start (preset or explicit config), profiler_stop (saves profile to downloads directory)
Screencast: screencast_start (records the page viewport to a video file in the downloads directory), screencast_stop (requires Firefox 154+)
Utilities: accept/dismiss dialog, history back/forward, set viewport
Saving bulky output to disk
Large tool output can consume significant context in CLI clients like Claude Code. The
screenshot_page, screenshot_by_uid, take_snapshot, list_console_messages,
list_network_requests, get_network_request, evaluate_script, and
evaluate_privileged_script tools accept an optional saveTo parameter that writes the
result to a file instead of returning it inline. saveTo takes one of three forms:
a file path (relative to the current working directory, or absolute within
~/.firefox-devtools-mcp; parent directories are created)an existing directory (a timestamped file is generated inside it)
true(a timestamped file is generated under~/.firefox-devtools-mcp/output/)
The response returns the path and byte size. The saved file always holds the full, untruncated data: the inline size safeguards (console message caps, network header truncation, snapshot line caps) never apply to it.
The text-producing tools (everything except the screenshots) also accept preview, a number
of characters of the saved output to echo back inline as a short excerpt. Screenshots have no
preview.
screenshot_page({ saveTo: "page.png" })
take_snapshot({ saveTo: true })
list_network_requests({ urlContains: "api", saveTo: "network.json" })
evaluate_script({ function: "() => performance.getEntries()", saveTo: true, preview: 2000 })By default, save paths are restricted: relative paths resolve against the current working
directory, and absolute paths are only allowed within ~/.firefox-devtools-mcp. Paths that
escape these locations are rejected. Start the server with --unrestricted-save-paths to
write to arbitrary locations, including absolute paths outside that directory.
Saved files can then be viewed with Claude Code's Read tool without impacting context size.
Local development
npm install
npm run build
# Run with Inspector against local build
npx @modelcontextprotocol/inspector node dist/index.js --headless --viewport 1280x720
# Or run in dev with hot reload
npm run inspector:devSee CONTRIBUTING.md for more details on local development, testing, and CI.
Troubleshooting
Firefox not found: pass
--firefox-path "/Applications/Firefox.app/Contents/MacOS/firefox"(macOS) or the correct path on your OS.First run is slow: Selenium sets up the BiDi session; subsequent runs are faster.
Stale UIDs after navigation: take a fresh snapshot (
take_snapshot) before using UID tools.Windows 10: Error during discovery for MCP server 'firefox-devtools': MCP error -32000: Connection closed
Solution 1 Wrap with
cmd /c(details):"mcpServers": { "firefox-devtools": { "command": "cmd", "args": ["/c", "npx", "-y", "@mozilla/firefox-devtools-mcp@latest"] } }Solution 2 Use the absolute path to
npx(adjust extension —.cmd,.bat,.exe, or.ps1— to match your setup):"mcpServers": { "firefox-devtools": { "command": "C:\\nvm4w\\nodejs\\npx.ps1", "args": ["-y", "@mozilla/firefox-devtools-mcp@latest"] } }
Versioning
Pre‑1.0 API: versions start at
0.x. Use@latestwith npx for the newest release.
Contributing
See CONTRIBUTING.md for how to file issues, run tests, and work on the project locally.
Author
Maintained by Mozilla.
License
Licensed under either of MIT or Apache 2.0 at your option.
Available Tools
31 toolsaccept_dialogA
Accept browser dialog. Provide promptText for prompts.
Inputs: promptText.
clear_downloadsA
Clear the tracked downloads buffer.
Inputs: no named parameters.
clear_snapshotA
Clear snapshot cache. Usually not needed.
Inputs: no named parameters.
click_by_uidA
Click element by UID. Set dblClick for double-click.
Inputs: uid (required), dblClick.
close_pageA
Close tab by index.
Inputs: pageIdx (required).
dismiss_dialogA
Dismiss browser dialog.
Inputs: no named parameters.
drag_by_uid_to_uidB
Drag element to another (HTML5 drag events).
Inputs: toUid (required), fromUid (required).
evaluate_scriptA
Execute JS function in page. Prefer UID tools for interactions.
Inputs: function (required), args, saveTo, preview, +1 more.
fill_by_uidB
Fill text input/textarea by UID.
Inputs: uid (required), value (required).
fill_form_by_uidA
Fill multiple form fields at once.
Inputs: elements (required).
get_firefox_infoARead-only
Get information about the current Firefox instance configuration, including binary path, environment variables, and output file location.
Inputs: no named parameters.
get_firefox_outputARead-only
Retrieve Firefox output (stdout/stderr including MOZ_LOG, warnings, crashes, stack traces). Returns recent output from the capture file. Use filters to focus on specific content.
Inputs: grep, lines, since.
hover_by_uidB
Hover over element by UID.
Inputs: uid (required).
install_extensionA
Install a Firefox extension using WebDriver BiDi webExtension.install command. Supports installing from archive (.xpi/.zip), base64-encoded data, or unpacked directory.
Inputs: type (required), path, value, permanent.
list_downloadsARead-only
List downloads tracked since startup, including status and saved file path.
Inputs: limit, format, status, urlContains.
list_pagesARead-only
List open tabs (index, title, URL). Selected tab is marked.
Inputs: no named parameters.
new_pageA
Open new tab at URL. Returns tab index.
Inputs: url (required).
resolve_uid_to_selectorARead-only
Resolve UID to CSS selector. Fails if stale.
Inputs: uid (required).
restart_firefoxA
Restart Firefox with different configuration. Allows changing binary path, environment variables, and other options. All current tabs will be closed.
Inputs: env, prefs, headless, startUrl, +2 more.
screencast_startA
Start recording a screencast (video) of the current page viewport, saving the output to a file in the downloads directory. Returns a screencast id to pass to screencast_stop. Multiple recordings can run at once.
Inputs: width, height, context, mimeType, +1 more.
screencast_stopA
Stop an in-progress screencast recording started with screencast_start and finalize the video file. Returns the path to the saved file.
Inputs: screencast.
screenshot_by_uidARead-only
Capture element screenshot by UID as base64 PNG.
Inputs: uid (required), saveTo.
screenshot_pageARead-only
Capture page screenshot as base64 PNG.
Inputs: saveTo.
select_pageA
Select active tab by index, URL, or title. Index takes precedence.
Inputs: url, title, pageIdx.
set_download_behaviorA
Control how downloads are handled: allow (save silently to the default download directory), deny (cancel), or reset to default. Avoids the native save-file dialog. Requires a recent Firefox.
Inputs: behavior (required).
set_viewport_sizeA
Set viewport dimensions in pixels.
Inputs: width (required), height (required).
take_snapshotARead-only
Capture DOM snapshot with stable UIDs. Retake after navigation.
Inputs: saveTo, preview, maxDepth, maxLines, +4 more.
uninstall_extensionA
Uninstall a Firefox extension using WebDriver BiDi webExtension.uninstall command. Requires the extension ID returned by install_extension or obtained from list_extensions.
Inputs: id (required).
upload_file_by_uidB
Upload file to file input by UID.
Inputs: uid (required), filePath (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 31 tools
Most tools target distinct resource+action pairs (navigate, close, list, select, click, hover, fill, drag, upload, etc.). A few near-overlaps exist, such as get_firefox_info vs get_firefox_output and screenshot_page vs screencast_start, but their descriptions clarify the difference.
The dominant pattern is verb_noun, often with a by_uid suffix for element interactions (click_by_uid, hover_by_uid, etc.). Minor inconsistencies include list_pages vs new_page (plural vs singular) and the longer resolve_uid_to_selector, but the overall convention is clear.
At 31 tools, this exceeds the 25+ threshold for a heavy tool set. While many tools cover distinct features, several are auxiliary helpers (clear_snapshot, resolve_uid_to_selector, get_firefox_info) that could potentially be merged or removed without losing core functionality.
The tool surface covers a broad range of browser automation needs: page lifecycle, element interaction, snapshots, screenshots, downloads, dialogs, history, viewport control, extensions, screencasts, and script execution. Minor gaps exist (no explicit wait/condition, no direct text extraction, no iframe handling), but agents can generally work around them.
Maintenance
Related MCP Connectors
Official Browserless MCP server: browser automation, scraping, screenshots and PDFs for AI agents.
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with yo…
A Model Context Protocol server for Wix AI tools
Model Context Protocol server for the Apideck Unified API. Connect any MCP-compatible agent framework to 100+ accounting systems, HRIS platforms, file storage providers, and more through one integration. More information https://www.apideck.com/mcp-server
Related MCP Servers
- AlicenseDqualityDmaintenanceAI-driven browser automation server that implements the Model Context Protocol to enable natural language control of web browsers for tasks like navigation, form filling, and visual interaction.12MIT
- AlicenseNot gradedqualityDmaintenanceA Model Context Protocol server implementation that enables browser automation through standardized MCP clients, supporting features like navigation, element interaction, and screenshots across Chrome, Firefox, and Edge browsers.1,042 npmMIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server that exposes Firefox browser automation capabilities through the Model Context Protocol, enabling AI assistants to control web browsers for navigation, interaction, and data collection tasks.4BSD 3-Clause
- AlicenseAqualityBmaintenanceMCP server for browser automation with anti-detection. Scout pages, find elements, interact with websites, and monitor network traffic from any AI client that supports the Model Context Protocol.211MIT