fire-fodex
This server enables an AI agent to remotely control Firefox browser tabs via MCP (Model Context Protocol) for comprehensive browser automation.
Tab Management: List all open tabs, open new tabs, claim existing tabs for agent control, release claimed tabs without closing, and close tabs entirely.
Navigation: Navigate to URLs, go back/forward in browsing history, and reload pages.
Page Inspection: Capture compact DOM snapshots (configurable element count/text length), take PNG screenshots (with cropping and full-page options), and find visible elements by text, role, label, placeholder, CSS selector, or auto-detection.
Synchronization: Wait for specific conditions such as text appearance, selector presence, navigation completion, URL changes, page load, or element focus.
Mouse & Pointer Interaction: Click by coordinates or selector, move the mouse, scroll pages or specific elements, and drag elements from one point to another.
Keyboard & Text Input: Type text into elements (with optional clear-first), paste text, send key presses, focus elements, and select text ranges in inputs, textareas, or contenteditable elements.
Script Execution: Run arbitrary JavaScript directly on the page for custom automation, with built-in protections blocking direct cookie/storage extraction.
All operations are session-based, requiring a session tab ID obtained by listing or claiming tabs.
Allows an AI agent to control Firefox, including listing open tabs, navigating, clicking, typing, scrolling, taking screenshots, and executing page-side actions via a WebExtension and native messaging.
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., "@fire-fodex list my open Firefox tabs"
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.
fire-fodex
Firefox を、MCP agent からそのまま触るためのローカルツールです。
Chrome じゃなくて Firefox を使いたい。
でも AI agent からタブを見たり、クリックしたり、スクショを取ったりしたい。
そのために WebExtension、native messaging、MCP server をつないでいます。
OpenAI の Chrome plugin の移植ではありません。Firefox 用に別で作った橋です。
まだ v1 なので、かなりローカル開発者向けです。拡張機能は手で読み込みます。MCP server も自分で起動します。普段使いのブラウザプロファイルで動かす前に、コードは一回見てください。
できること
Firefox のタブ一覧を見る
新しいタブを開く
既存タブを agent 用に確保する
ページ移動、リロード、戻る、進む
表示中ページの軽い snapshot を取る
画面内の文字を探す
クリック、入力、貼り付け、フォーカス、テキスト選択、キー入力
マウス移動、スクロール、ドラッグ
文字や selector が出るまで待つ
通常スクショと full-page スクショ
Related MCP server: firefox-bridge
やらないこと
Firefox の cookie、保存済みパスワード、profile database、localStorage は直接読みません
ブラウザの情報を勝手に外へ送りません
cloud sync、account auth、操作録画、権限 UI はありません
AMO 向けの配布はまだしていません
firefox_eval はあります。ページ側で少しだけ処理したい時の逃げ道です。
ただし sandbox ではありません。cookie、storage、password 抜き取りっぽい文字列は止めますが、それで安全と言い切れるものではないです。信用できない agent に、普段の Firefox profile を触らせないでください。
必要なもの
Windows
Firefox Stable
Node.js 24+
npm 11+
セットアップ
cd path\to\fire-fodex
npm install
npm run build
npm run install:hostFirefox 側で拡張機能を読み込みます。
Firefox を開く
about:debugging#/runtime/this-firefoxに行くLoad Temporary Add-on...を押すextension\manifest.jsonを選ぶ
MCP server を起動します。
npm run mcpMCP client 側の設定例です。
{
"mcpServers": {
"fire-fodex": {
"command": "node",
"args": [
"path\\to\\fire-fodex\\dist\\mcp-server\\index.js"
]
}
}
}Tools
firefox_list_tabsfirefox_claim_tabfirefox_open_tabfirefox_navigatefirefox_snapshotfirefox_screenshotfirefox_findfirefox_wait_forfirefox_mouse_clickfirefox_mouse_movefirefox_scrollfirefox_dragfirefox_focusfirefox_select_textfirefox_pastefirefox_evalfirefox_clickfirefox_typefirefox_keypressfirefox_backfirefox_forwardfirefox_reloadfirefox_close_tabfirefox_release
基本は、見てから触る流れです。
firefox_snapshotやfirefox_findで画面を見るform は
firefox_focus、firefox_paste、firefox_type、firefox_keypresspointer 操作は
firefox_mouse_click、firefox_mouse_move、firefox_scroll、firefox_dragページが変わる操作の後は
firefox_wait_for見た目が大事なら
firefox_screenshot
確認
npm run typecheck
npm test
npm run lint:extension
npm run check
powershell -ExecutionPolicy Bypass -File scripts\smoke.ps1smoke.ps1 は、native host を入れて Firefox 側で拡張機能を読み込んでから動かしてください。
権限
拡張機能は <all_urls> を要求します。普通のサイトを横断して操作するためです。
強い権限です。v1 は「中身を読んで、自分のローカルで動かす」人向けとして置いています。
仕組み
AI agent / MCP client
|
| stdio MCP
v
fire-fodex MCP server
|
| localhost JSON request
v
native messaging host
|
| Firefox native messaging stdio
v
Firefox WebExtension
|
| tabs + content scripts
v
Firefox tabs拡張機能がつながっている間、native host は 127.0.0.1:17365 で待ち受けます。
Available Tools
24 toolsfirefox_backC
Go back in a Firefox tab.
Inputs: sessionTabId (required).
firefox_claim_tabC
Claim an existing Firefox tab by sessionTabId.
Inputs: sessionTabId (required).
firefox_clickC
Click an element in a Firefox tab.
Inputs: sessionTabId (required), selector (required).
firefox_close_tabC
Close a Firefox tab.
Inputs: sessionTabId (required).
firefox_dragC
Drag from one point or element to another.
Inputs: sessionTabId (required), fromX, fromY, fromSelector, +6 more.
firefox_evalB
Run page JavaScript for local browser automation. Obvious cookie/storage extraction is blocked.
Inputs: sessionTabId (required), script (required), frameRef.
firefox_findB
Find visible page elements by text, role, label, placeholder, CSS selector, or approximate query.
Inputs: sessionTabId (required), query (required), by, frameRef, +1 more.
firefox_focusC
Focus an element by selector/ref or viewport point.
Inputs: sessionTabId (required), selector, x, y, +2 more.
firefox_forwardB
Go forward in a Firefox tab.
Inputs: sessionTabId (required).
firefox_keypressC
Send a keyboard key to a Firefox tab.
Inputs: sessionTabId (required), key (required).
firefox_list_tabsA
List current Firefox tabs.
Inputs: no named parameters.
firefox_mouse_clickC
Click by viewport coordinates or selector/ref.
Inputs: sessionTabId (required), x, y, selector, +4 more.
firefox_mouse_moveC
Move the mouse over viewport coordinates or an element.
Inputs: sessionTabId (required), x, y, selector, +2 more.
firefox_open_tabC
Open a new Firefox tab.
Inputs: url, active.
firefox_pasteC
Paste text into the focused element or a chosen element.
Inputs: sessionTabId (required), text (required), selector, frameRef, +1 more.
firefox_releaseB
Release a claimed Firefox tab without closing it.
Inputs: sessionTabId (required).
firefox_reloadC
Reload a Firefox tab.
Inputs: sessionTabId (required).
firefox_screenshotC
Capture a PNG screenshot of a Firefox tab.
Inputs: sessionTabId (required), fullPage, cropX, cropY, +2 more.
firefox_scrollC
Scroll the page or a specific element.
Inputs: sessionTabId (required), deltaX, deltaY, selector, +2 more.
firefox_select_textC
Select text in an input, textarea, contenteditable element, or the current document.
Inputs: sessionTabId (required), selector, start, end, +2 more.
firefox_snapshotC
Read a compact DOM snapshot from a Firefox tab.
Inputs: sessionTabId (required), maxTextLength, maxElements.
firefox_typeC
Type text into an element in a Firefox tab.
Inputs: sessionTabId (required), selector (required), text (required), clear.
firefox_wait_forC
Wait for text, selector, navigation, URL substring, load completion, or focus state.
Inputs: sessionTabId (required), kind (required), value, frameRef, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 24 tools
Most tools have distinct purposes, but firefox_click and firefox_mouse_click overlap in functionality, as both can click using selectors. Additionally, firefox_focus and firefox_click could be confused. Overall, the distinctions are clear enough for most use cases.
All tools consistently use the 'firefox_' prefix. The naming pattern is mostly verb or verb_noun, but there are exceptions like 'mouse_click' (noun+verb) and 'mouse_move', which break the pattern slightly. The majority follow a predictable style.
24 tools is on the high side for a browser automation server, but each tool covers a specific action. While the count is above the typical 3-15 range, it is not excessive and reflects the complexity of controlling a browser.
The tool set covers essential browser automation tasks: navigation, tab management, user input, scrolling, screenshots, and DOM interaction. Missing features like alert handling or file uploads are minor gaps, but core workflows are well-supported.
Maintenance
Related MCP Connectors
Stealth web browser for agents: search, fetch, click, download and type in persistent MCP sessions.
Web scraping, browser automation, screenshots and AI data extraction for AI agents via MCP.
Official Browserless MCP server: browser automation, scraping, screenshots and PDFs for AI agents.
Unofficial Browserless MCP — wraps the Browserless headless-Chromium REST API (browserless.io)
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceA local MCP server for Firefox that gives AI agents full control over the browser via a Unix socket, enabling automation of tabs, pages, cookies, and more without exposing any network ports.MIT
- AlicenseNot gradedqualityBmaintenanceEnables MCP-capable CLIs to operate real, already-logged-in Firefox tabs via a WebExtension and native messaging, without simulated input. Supports navigation, clicking, typing, reading, screenshots, and console/network capture with policy gating and frame awareness.2MIT
- AlicenseNot gradedqualityBmaintenanceMCP server that gives local clients controlled access to a Firefox profile, enabling tab listing, reading, interaction, scripting, and screenshots with granular, revocable permissions.1MIT
- AlicenseNot gradedqualityBmaintenanceEnables an MCP client to drive a real, already-logged-in Firefox session through a WebExtension rather than a WebDriver connection, so tabs, cookies and logins work as-is without relaunching the browser under remote-control flags. Exposes tools for tab listing and selection, navigation, accessibility-style snapshots, clicking, filling, typing, key presses, file uploads, screenshots and waiting, with per-origin permission grants and opt-in script evaluation.MIT