Termux Browser Pilot

Real browser automation for Termux/Android. No root required.

Firefox (default) or Chromium on Xvfb — runs entirely on your phone. Firefox passes Cloudflare natively via TLS fingerprint. Chromium uses CDP stealth. Both use real device hardware info.

Features

• Persistent daemon — browser stays running, sub-second commands
• Passes Cloudflare — Firefox: instant TLS-based pass; Chromium: Turnstile handler
• No automation framework — Firefox controlled via xdotool + clipboard JS execution
• AI-friendly — --json flag on all commands for structured output
• Real device fingerprint — auto-detects screen, GPU, CPU, arch, model
• Human-like input — Bezier curve mouse, realistic typing, clipboard paste
• Session persistence — auto-save/load cookies across restarts
• Cookie management — get/set/save/load, Netscape format import/export
• Network tracking — capture all request URLs, types, sizes (Chromium only)
• Screenshots — viewport or full page PNG, PDF export
• Accessibility tree — ARIA roles and names
• Cross-device — auto-adapts to any Android device's real specs
• Smart element targeting — find clickable elements by visible text
• Idle timeout — auto-shutdown daemon after configurable inactivity
• Tab management — open, close, switch between browser tabs
• Proxy support — HTTP/SOCKS5 proxy for Firefox and Chromium
• Request interception — block URLs/domains via JS fetch/XHR hooks
• Multi-step macros — chain commands into reusable JSON scripts
• Console log capture — capture console.log/warn/error/info output
• Download management — auto-download files to ~/.tbp/downloads/
• Network request log — capture URLs, types, durations, sizes
• DOM mutation observer — watch for element changes in real-time
• Element screenshot — capture specific elements by CSS selector
• Drag and drop — smooth mouse drag between elements or by offset
• Iframe support — list, eval, text, click inside same-origin iframes
• File upload — set files on input[type=file] via JS DataTransfer
• Geolocation spoofing — override navigator.geolocation coordinates
• User agent switching — optional override (default: real browser UA)
• Cookie injection — set cookies with domain, path, secure, SameSite
• Local/session storage — get, set, remove, clear, list items
• Clipboard access — read/write Xvfb system clipboard
• Form auto-fill — fill all form fields from JSON spec
• Lightweight — no Puppeteer/Playwright dependency

Quick Start

Install

bash setup.sh

# Or manually:
pkg install tur-repo x11-repo
pkg install firefox xorg-server-xvfb xdotool xclip openbox python3
pip install websockets  # only needed for Chromium mode

CLI (Daemon Mode — Recommended)

Browser starts once, stays running. Commands are instant.

# Navigate (auto-starts daemon on first command)
tbp goto https://example.com
tbp goto https://audiogames.net -cf    # Cloudflare bypass

# Read content
tbp text                               # Page text
tbp links                              # All links
tbp eval "document.title"              # Run JavaScript
tbp a11y                               # Accessibility tree

# Interact
tbp click "button.submit"              # Click element
tbp type "input[name=q]" "hello"       # Type into field
tbp press Enter                        # Press key
tbp scroll --down                      # Scroll page
tbp find "Sign In"                     # Find elements by text
tbp find "Submit" --role button        # Filter by role

# Tab management
tbp tab new                            # Open new tab
tbp tab new https://example.com        # New tab with URL
tbp tab close                          # Close current tab
tbp tab next                           # Next tab
tbp tab prev                           # Previous tab
tbp tab goto 2                         # Switch to tab 2

# Block requests
tbp block ads.example.com tracker.js   # Block URL patterns
tbp blocklist                          # List blocked patterns
tbp unblock ads.example.com            # Unblock pattern

# Macros (JSON scripts)
tbp macro workflow.json                # Run multi-step macro

# Console log capture
tbp console start                      # Start capturing logs
tbp console stop                       # Stop capturing logs
tbp console logs                       # Show captured logs
tbp console logs --clear               # Show and clear buffer
tbp console clear                      # Clear log buffer

# Downloads
tbp downloads                          # List downloaded files

# Network request log
tbp network start                      # Start logging requests
tbp network stop                       # Stop logging requests
tbp network logs                       # Show captured requests
tbp network logs --clear               # Show and clear
tbp network clear                      # Clear request log

# DOM mutation observer
tbp observe start                      # Start watching mutations
tbp observe stop                       # Stop watching mutations
tbp observe logs                       # Show captured mutations
tbp observe clear                      # Clear mutations

# Element screenshot
tbp screenshot-element "div.hero"      # Screenshot specific element
tbp screenshot-element "#logo" logo.png

# Drag and drop
tbp drag ".item" --target ".dropzone"  # Drag to element
tbp drag ".slider" --dx 100            # Drag by offset

# Iframe support
tbp iframe list                        # List all iframes
tbp iframe eval "iframe#f1" "document.title"  # JS in iframe
tbp iframe text "iframe#f1"            # Get iframe text
tbp iframe click "iframe#f1" "button"  # Click in iframe

# File upload
tbp upload "input[type=file]" photo.jpg  # Upload file (max 5MB)

# Geolocation spoofing
tbp geo set 40.7128 -74.0060           # Set to New York
tbp geo set 51.5074 -0.1278 -a 50      # London, 50m accuracy
tbp geo clear                           # Clear override

# User agent switching (default: real browser UA — override is optional)
tbp useragent set "Mozilla/5.0 (Windows NT 10.0; Win64; x64)"
tbp ua clear                            # Restore real UA
# Note: Overrides JS-side navigator.userAgent only.
# HTTP User-Agent header stays real (requires restart to change).

# Cookie injection
tbp cookie-set session_id abc123          # Simple cookie
tbp cs token xyz --domain .example.com    # With domain
tbp cs pref dark --max-age 86400 --secure # Secure + 1 day

# Local/session storage
tbp storage list                          # List localStorage
tbp storage get mykey                     # Get value
tbp storage set mykey myvalue             # Set value
tbp storage remove mykey                  # Remove key
tbp storage clear                         # Clear all
tbp storage list --type session           # sessionStorage

# Clipboard
tbp clipboard read                        # Read clipboard
tbp clip write "Hello world"              # Write to clipboard

# Form auto-fill (JSON file: [{selector, value}, ...])
tbp form-fill form_data.json              # Fill all fields

# Proxy support (on start)
tbp start --proxy socks5://127.0.0.1:1080

# Screenshots
tbp screenshot page.png
tbp screenshot page.png --full         # Full page

# AI-friendly JSON output
tbp goto https://example.com --json
tbp text --json
tbp links --json

# Session management
tbp status                             # Daemon status
tbp cookies --save session.json        # Save cookies
tbp cookies --load session.json        # Load cookies
tbp stop                               # Graceful shutdown

Python API

import asyncio
from src.pilot import Pilot

async def main():
    # Firefox (default) — passes Cloudflare natively
    async with Pilot() as p:
        await p.goto_cf("https://audiogames.net")
        print(await p.title())
        await p.screenshot("page.png")

    # Chromium mode with session persistence
    async with Pilot(browser="chromium", session_file="session.json") as p:
        await p.goto("https://example.com")
        await p.type("input[name=q]", "hello")

asyncio.run(main())

Daemon Client (Python)

from src.client import send_command
import asyncio

async def main():
    # Auto-starts daemon if not running
    r = await send_command("goto", {"url": "https://example.com"})
    print(r["data"]["title"])

    r = await send_command("text")
    print(r["data"]["text"])

    r = await send_command("screenshot", {"path": "shot.png"})

asyncio.run(main())

MCP Server (for Claude Code / AI Agents)

# Register with Claude Code
claude mcp add tbp -- tbp-mcp

# Or use .mcp.json (already included in repo)
# Tools: browser_goto, browser_text, browser_click, browser_type,
#        browser_screenshot, browser_a11y, browser_links, browser_eval,
#        browser_scroll, browser_hover, browser_press, browser_find,
#        browser_tab_*, browser_block, browser_unblock, browser_blocklist,
#        browser_macro, browser_wait, browser_wait_for, browser_cookies_*,
#        browser_console_*, browser_downloads, browser_network_*,
#        browser_observe_*, browser_mutations*, browser_screenshot_element,
#        browser_drag, browser_iframe_*, browser_upload, browser_geo_*,
#        browser_useragent_*, browser_cookie_set, browser_storage_*,
#        browser_clipboard_*, browser_form_fill, browser_status

Requires: pip install "mcp[cli]>=1.0" (or pip install termux-browser-pilot[mcp]).

Feature Reference

Navigation & Content

Command	Description
goto URL [-cf]	Navigate to URL. -cf for Cloudflare bypass
back / forward / reload	History navigation
text [--selector S]	Get visible page text
html [--selector S]	Get page HTML
links [--limit N]	List all links
eval EXPRESSION	Execute JavaScript
title / url	Get page title/URL
find TEXT [--role R]	Find interactive elements by visible text
a11y	Accessibility tree (ARIA roles and names)

Interaction

Command	Description
click SELECTOR [--human]	Click element. --human for Bezier curve
type SELECTOR TEXT	Type text into input
press KEY	Press keyboard key (Enter, Tab, Escape...)
scroll [--up] [--amount N]	Scroll page
hover SELECTOR	Hover over element
drag SOURCE [--target T] [--dx N --dy N]	Drag element

Tab Management

Command	Description
tab new [URL]	Open new tab
tab close	Close current tab
tab next / tab prev	Switch tabs
tab goto INDEX	Switch to tab 1-9

Screenshots & Export

Command	Description
screenshot [PATH] [--full]	Screenshot (viewport or full page)
screenshot-element SELECTOR [PATH]	Screenshot specific element
pdf [PATH]	Export page as PDF

Request Blocking

Command	Description
block PATTERNS...	Block URL patterns (domains/substrings)
unblock PATTERNS...	Remove blocked patterns
blocklist	List current blocked patterns

Console Log Capture

Command	Description
console start	Start capturing console.log/warn/error/info
console stop	Stop capturing (disables re-injection)
console logs [--clear] [--limit N]	Show captured logs
console clear	Clear log buffer

Network Request Log

Uses PerformanceObserver (passive, no conflict with URL blocker).

Command	Description
network start	Start logging network requests
network stop	Stop logging
network logs [--clear] [--limit N]	Show requests (URL, type, duration, size)
network clear	Clear request log

DOM Mutation Observer

Command	Description
observe start	Start watching DOM mutations
observe stop	Stop watching
observe logs [--clear] [--limit N]	Show mutations (childList, attributes, etc.)
observe clear	Clear mutation buffer

Iframe Support

Works with same-origin iframes only. Cross-origin iframes are listed but inaccessible.

Command	Description
iframe list	List all iframes (index, src, accessibility)
iframe eval SELECTOR EXPRESSION	Execute JS inside iframe
iframe text SELECTOR [--inner S]	Get text from iframe
iframe click SELECTOR TARGET	Click element inside iframe

File Upload

Sets files on <input type="file"> elements via JS DataTransfer API. Max 5MB.

Command	Description
upload SELECTOR PATH	Upload file to file input

Geolocation Spoofing

Overrides navigator.geolocation.getCurrentPosition and watchPosition. Re-injected automatically after page navigation. Saves/restores originals on clear.

Command	Description
geo set LAT LNG [-a ACCURACY]	Set geolocation (accuracy default: 100m)
geo clear	Restore real geolocation behavior

User Agent Switching

Default: real browser user agent. Override is optional and JS-side only (navigator.userAgent). HTTP User-Agent header sent to servers stays real. Re-injected after navigation. Restored on clear.

Command	Description
useragent set UA_STRING	Override navigator.userAgent
useragent clear / ua clear	Restore real user agent

Cookie Injection

Sets cookies via document.cookie. Cannot set HttpOnly cookies (browser security). Values are URL-encoded automatically. Name must not contain =, ;, or newlines.

Command	Description
cookie-set NAME VALUE [--domain D] [--path P]	Set cookie
Options: --max-age N, --secure, --samesite Strict|Lax|None

Local/Session Storage

Command	Description
storage list [--type local|session]	List all items
storage get KEY [--type ...]	Get value by key
storage set KEY VALUE [--type ...]	Set key-value pair
storage remove KEY [--type ...]	Remove key
storage clear [--type ...]	Clear all items

Clipboard Access

Reads/writes the Xvfb virtual display clipboard (not the phone clipboard). Useful for reading after a page copy, or setting before a Ctrl+V paste.

Command	Description
clipboard read	Read clipboard text
clipboard write TEXT	Write text to clipboard

Form Auto-fill

Fills multiple fields from a JSON file. Handles text inputs, selects, checkboxes, and radios. Dispatches input + change events (React/Vue compatible).

[
  {"selector": "#name", "value": "John Doe"},
  {"selector": "#email", "value": "[email protected]"},
  {"selector": "select#country", "value": "US"},
  {"selector": "#agree", "value": true}
]

Command	Description
form-fill FILE / ff FILE	Fill fields from JSON file (max 100)

CSS Injection

Inject, remove, and list custom stylesheets. Persists across scrolling and re-injects after navigation.

Command	Description
css inject "RULES" [--id NAME]	Inject CSS (text or file path)
css remove [--id NAME]	Remove stylesheet (or all)
css list	List injected stylesheets

Example: tbp css inject "body{background:#111;color:#eee}" --id darkmode

Wait+Action

Wait for an element to appear in the DOM, then automatically perform an action. Uses MutationObserver for efficient detection.

Command	Description
waitact SELECTOR [--click]	Wait then click (default)
waitact SELECTOR --type TEXT	Wait then type
waitact SELECTOR --text	Wait then get text

Timeout: Default 10s, max 120s. Uses --timeout flag.

Page Event Capture

Capture DOM events (click, submit, input, change, keydown) with target element info. Re-injects after navigation automatically.

Command	Description
events start [--types click submit ...]	Start capturing events
events stop	Stop and remove listeners
events logs [--limit N] [--clear]	Show captured events
events clear	Clear event log

Viewport/Window Resize

Control browser window dimensions for responsive testing.

Command	Description
viewport set WIDTH HEIGHT / vp set	Resize window
viewport get / vp get	Get current dimensions

Returns both window size (xdotool) and inner viewport (JS). Range: 100-7680 x 100-4320.

Page Search

Find text on the page with visual highlighting. Navigate between matches.

Command	Description
search find "text" [--case-sensitive]	Find and highlight all matches
search next	Go to next match
search prev	Go to previous match
search clear	Clear highlights

Highlights persist across scrolling. Current match shown in orange, others in yellow. Re-injects after navigation if search is active.

Shadow DOM Access

Query, read, and click elements inside shadow DOM boundaries. Recursively traverses all shadow roots to find elements.

Command	Description
shadow query SELECTOR	Find element (returns tag, id, text, attributes)
shadow text SELECTOR	Get text content from shadow element
shadow click SELECTOR	Click element inside shadow root

Response Body Capture

Capture fetch/XHR response bodies for API inspection and debugging. Bodies truncated to 10KB per response, max 500 entries.

Command	Description
responses start	Start capturing response bodies
responses stop	Stop capturing
responses logs [--limit N] [--clear]	Show captured responses
responses clear	Clear response log

Multi-tab Session Management

Save and restore all open browser tabs as named sessions. Sessions stored in ~/.tbp/sessions/{name}.json.

Command	Description
session save NAME	Save all open tabs
session load NAME	Restore tabs from session
session list	List saved sessions
session delete NAME	Delete a session

Note: Tab detection uses Ctrl+1-9 shortcuts (max 9 tabs).

HTTP Header Injection

Inject custom headers into all fetch() and XMLHttpRequest calls. Headers persist across requests and re-inject after navigation. Useful for auth tokens, API keys, or custom tracking headers.

Command	Description
headers set Name:Value [...]	Set one or more custom headers
headers list	List active custom headers
headers clear	Remove all custom headers and restore original fetch/XHR

Note: Headers are injected JS-side only (via fetch/XHR interception). They do not affect initial page load requests or non-JS resource fetches.

Page Performance Metrics

Get timing and resource metrics from the Performance API.

Command	Description
perf	Show page performance metrics

Returns: first byte time, DOM interactive, DOM content loaded, load complete, DOM element count, script/stylesheet/image/iframe counts, JS heap memory usage.

Element Attributes

Read, set, or remove HTML attributes on elements.

Command	Description
attr get SELECTOR [--name NAME]	Get one or all attributes
attr set SELECTOR NAME VALUE	Set an attribute
attr remove SELECTOR NAME	Remove an attribute

Browser Profile Management

Save and restore browser state (cookies + localStorage + URL) as named profiles. Profiles are stored in ~/.tbp/profiles/<name>/.

Command	Description
profile save NAME	Save current state as named profile
profile load NAME	Load saved profile (cookies + localStorage)
profile list	List all saved profiles
profile delete NAME	Delete a saved profile

Profile names: Alphanumeric, hyphens, and underscores only.

Element Highlight

Visual debugging — outline matching elements with colored borders.

Command	Description
highlight add SELECTOR [--color red] [--label TEXT]	Highlight elements
highlight clear [SELECTOR]	Remove highlights (all or specific)

Highlights persist across navigation (re-injected after goto).

PDF Export Options

Extended PDF export with layout control.

Option	Description
--landscape	Landscape orientation
--scale 0.1-2.0	Scale factor
--margin-top/right/bottom/left N	Margins in inches
--page-ranges '1-3'	Specific pages
--no-background	Skip background graphics

Example: tbp pdf report.pdf --landscape --scale 0.8 --margin-top 0.5

Cookie Auth Sessions

Save/restore authentication via cookies — auto-navigates to saved URL.

Command	Description
auth save NAME	Save cookies + URL as auth session
auth load NAME	Restore cookies and navigate to saved URL
auth list	List saved auth sessions
auth delete NAME	Delete auth session

Sessions stored in ~/.tbp/auth/ with 0o600 permissions.

Network Throttling

Simulate slow connections by adding latency to fetch/XHR requests.

Command	Description
throttle set --preset 3g	Apply throttle preset
throttle set --latency 500	Custom latency (ms)
throttle clear	Remove throttling
throttle get	Show current config

Presets: 3g (400ms), slow-3g (2000ms), fast-3g (150ms), offline (reject all).

Annotated Screenshot (AI-First)

Screenshot with numbered badges on interactive elements + legend mapping numbers to CSS selectors.

Command	Description
annotate [--path FILE] [--selector CSS] [--max N]	Labeled screenshot

AI workflow: annotate → read legend → click using selector from legend.

Page Audit

One-command structured page health report.

Command	Description
audit	Full page health report

Returns: elements, links (total/external/broken), images (alt/broken), forms, headings, meta, scripts, page size, load time, console errors.

Response Mocking

Return fake responses for matching URL patterns. Useful for testing without real backends.

Command	Description
mock set PATTERN BODY [--status N] [--content-type CT]	Add mock
mock clear [PATTERN]	Remove mock(s)
mock list	List active mocks

Mocks persist across navigation (re-injected after goto). Max 50 mocks.

DOM Snapshot & Diff

Save page state, perform actions, then diff to verify changes.

Command	Description
snapshot take NAME	Capture page state
snapshot diff NAME1 NAME2	Compare two snapshots
snapshot list	List snapshots
snapshot delete NAME	Delete snapshot

AI workflow: snapshot take before → actions → snapshot take after → snapshot diff before after.

Double-click

Command	Description
dblclick SELECTOR / dbl	Double-click element (dispatches dblclick event)
dblclick SELECTOR --human	Use xdotool double-click

Select Dropdown

Command	Description
select SELECTOR --value V	Select option by value
select SELECTOR --label L	Select option by visible text
select SELECTOR --index N	Select option by index

Dispatches change + input events for React/Vue compatibility.

Check / Uncheck

Command	Description
check SELECTOR	Check a checkbox or radio
check SELECTOR --action uncheck	Uncheck
check SELECTOR --action toggle	Toggle

Input Value

Command	Description
input-value SELECTOR / iv	Read current value of input/select/textarea

Returns {value, tag, type}.

Element State

Command	Description
element-state SELECTOR / es	Query element visibility/enabled/checked state

Returns {visible, enabled, checked, editable, tag, type}.

Bounding Box

Command	Description
bounding-box SELECTOR / bb	Get element position and dimensions

Returns {x, y, width, height, top, left, bottom, right, scrollX, scrollY}.

Scroll To Element

Command	Description
scroll-to SELECTOR / st	Scroll element into view (default: center)
scroll-to SELECTOR --block start	Align to top

Block options: center, start, end, nearest.

Set Page Content

Command	Description
set-content HTML	Replace page with raw HTML (no navigation)
set-content @file.html	Load HTML from file

Dialog Handling (alert/confirm/prompt)

Auto-handle browser dialogs without blocking. Monkey-patches window.alert/confirm/prompt. Re-injects after navigation.

Command	Description
dialog handle	Enable auto-accept for dialogs
dialog handle --reject	Auto-dismiss dialogs
dialog handle --prompt-text "answer"	Set prompt response
dialog dismiss	Shortcut for reject mode
dialog logs [--limit N] [--clear]	Show captured dialog messages
dialog clear	Clear dialog log

Wait for Response

Wait for a fetch/XHR response matching a URL pattern.

Command	Description
waitfor-response PATTERN / wr	Wait for matching response
waitfor-response PATTERN --timeout 30	Custom timeout (1-120s)

Enables response capture automatically if not already active.

Multi-step Macros

Chain commands into reusable JSON scripts (max 100 steps).

Command	Description
macro FILE	Run macro from JSON file

Session & Lifecycle

Command	Description
start [--proxy URL] [--idle-timeout N]	Start daemon
stop	Graceful shutdown
status	Show daemon status (PID, URL, uptime)
cookies [--save F] [--load F] [--clear]	Manage cookies
downloads / dl	List downloaded files
kill	Force-kill everything

Design Philosophy

Be real, not fake. Firefox runs as a normal browser — no geckodriver, no Marionette, no WebDriver flags. Controlled via xdotool (X11 native input) and clipboard-based JS execution through the Web Console. Chromium uses real device hardware info with only webdriver=false hidden.

Architecture

CLI (cli.py) ──Unix socket──→ Daemon (daemon.py)
                                ├── Pilot (unified API)
                                │    ├── BrowserPilot        Xvfb + openbox WM + browser lifecycle
                                │    ├── NativeFirefoxSession xdotool + clipboard JS (persistent profile)
                                │    ├── CDPSession          WebSocket CDP (Chromium)
                                │    ├── PageCommands        navigate, eval, text/html/links
                                │    ├── InputCommands       click, type, scroll, Bezier mouse
                                │    ├── ScreenshotCommands  PNG, full-page, PDF
                                │    ├── CookieCommands      get/set/save/load/export/import
                                │    ├── NetworkTracker      request capture (Chromium only)
                                │    ├── Accessibility       a11y tree
                                │    └── CloudflareHandler   Turnstile (Chromium only)
                                ├── DeviceInfo              auto-detect screen/GPU/CPU/model
                                └── Stealth                 CDP-level UA/platform (Chromium)

Auto-Detected Device Info

Property	Source	Example
Model	getprop ro.product.model	SM-S918B
Screen	DPI ratio + model DB	1080x2316
GPU	/sys/kernel/gpu/gpu_model	Adreno740v2
CPU cores	os.cpu_count()	8
RAM	/proc/meminfo	11GB
Architecture	uname -m	aarch64

Test Results (2026-02-24)

All 5 test sites pass with Firefox (default):

Site	Result	Notes
example.com	PASS	Navigation, JS eval, links, text, scroll
bot.sannysoft.com	PASS	Navigation, scroll
audiogames.net	PASS	CF bypassed via TLS fingerprint
nowsecure.nl	PASS	CF bypassed via TLS fingerprint
platform.openai.com	PASS	Form filling, clipboard paste input

Firefox vs Chromium

Feature	Firefox	Chromium
Cloudflare	Instant (TLS)	Turnstile handler
audiogames.net	PASS	FAIL
Network tracking	Not available	Full CDP tracking
JS execution	Console + clipboard	CDP Runtime.evaluate
Stealth needed	None	UA + WebGL + canvas

File Structure

termux-browser-pilot/
├── cli.py                 # CLI entry point (daemon + legacy commands)
├── setup.sh               # One-command installer
├── pyproject.toml          # Python packaging
├── src/
│   ├── pilot.py           # Main unified API + session persistence
│   ├── daemon.py          # Persistent browser daemon (Unix socket)
│   ├── client.py          # Daemon client (send commands)
│   ├── native.py          # Firefox: xdotool + clipboard JS execution
│   ├── browser.py         # Xvfb + browser lifecycle
│   ├── cdp.py             # CDP WebSocket client (Chromium)
│   ├── commands.py        # Page commands (navigate, eval, text, html)
│   ├── input.py           # Keyboard/mouse + Bezier human-like input
│   ├── screenshot.py      # Screenshot/PDF capture
│   ├── stealth.py         # Anti-detection (Chromium only)
│   ├── cookies.py         # Cookie management + Netscape import/export
│   ├── cloudflare.py      # CF Turnstile handler (Chromium)
│   ├── network.py         # Network tracking (Chromium only)
│   ├── accessibility.py   # A11y tree access
│   ├── device.py          # Real device detection
│   ├── gpu.py             # WebGL renderer spoofing
│   ├── lock.py            # Session lock (single-instance)
│   ├── _utils.py          # Shared utilities (JS escaping, binary checks)
│   └── mcp_server.py      # MCP server for Claude Code integration
└── .mcp.json              # Claude Code auto-discovery config

Requirements

• Termux on Android (aarch64)
• Firefox mode: firefox, xorg-server-xvfb, xdotool, xclip, openbox, python3
• Chromium mode: chromium, xorg-server-xvfb, mesa, python3, websockets>=12.0
• openbox (lightweight WM) is required for reliable keyboard shortcut routing in Xvfb

Troubleshooting

SSL Certificate Errors (Firefox)

If Firefox shows "Warning: Potential Security Risk Ahead" on HTTPS sites, it means Firefox isn't using Termux's system CA certificates. The daemon automatically fixes this on startup (v0.17.2+), but if you're on an older version or have an existing profile, add this to ~/.tbp/firefox_profile/user.js:

user_pref("security.enterprise_roots.enabled", true);

Then restart the daemon:

tbp stop
tbp goto https://example.com   # auto-starts with fix applied

Make sure the ca-certificates package is installed:

pkg install ca-certificates

Daemon Won't Start

If the daemon fails to start, check the log:

cat ~/.tbp/daemon.log

Common fixes:

• Kill stale processes: pkill -f Xvfb; pkill -f firefox; rm ~/.tbp/daemon.pid ~/.tbp/daemon.sock
• Ensure Xvfb is installed: pkg install xorg-server-xvfb

License

MIT