- Create comprehensive MCP Server documentation at docs/customize/mcp-server.mdx
- Add Integration group to navigation in docs.json
- Cover installation, Claude Desktop integration, available tools, and examples
- Address GitHub issue #2933 for missing MCP Server documentation
Auto-generated PR for: qwen support
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Adds first-class Qwen support via OpenAI-compatible API, with a working
example using qwen-vl-max and setup docs. Also tightens action schema
examples in prompts to improve output consistency.
- New Features - New features added
- Qwen support with example (examples/models/qwen.py) using qwen-vl-max,
vision enabled, and DashScope-compatible base_url.
- Documentation updates: Qwen guide, required ALIBABA_CLOUD env var, and
warning about smaller models returning incorrect action schemas.
- Prompt updates: replace generic action placeholder with a concrete
go_to_url example to enforce the correct nested schema.
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Prepare 0.7.4 release by updating the project version in pyproject.toml. No other changes.
<!-- End of auto-generated description by cubic. -->
Auto-generated PR for: ignore asci for chinese letters
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Preserves Chinese and other non-ASCII characters in extracted_content by
serializing JSON with ensure_ascii=False. Also updates .gitignore to
ignore Markdown files (*.md).
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Prevents syncing any events when the user isn’t authenticated and adds a
simple CLI auth flow so users can sign in and view runs in Browser Use
Cloud. Addresses ENG-475 by enforcing “no data without sign-in” and
providing a clear path to authenticate.
- **New Features**
- Added `browser-use auth` (device flow) that opens a session/task and
shows the Cloud URL.
- Added `Agent.authenticate_cloud_sync()` to sign in after a run.
- Introduced `UpdateAgentSessionEvent` and updated docs for setup and
reset.
- **Bug Fixes**
- CloudSync now skips sending events when not authenticated; only sends
during the auth flow or when signed in. Uses real user_id only after
auth and temp ID otherwise.
- Starts background auth at session creation; allows session/auth events
during auth to preserve first-run data, otherwise drops them.
- `DeviceAuthClient` accepts empty session IDs; clearing auth now
deletes the config file.
- Added a brief delay before first task to ensure the session exists.
<!-- End of auto-generated description by cubic. -->
Auto-generated PR for branch: fix-downloads-dedup-clean
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Fixes duplicate local download events by ensuring each download is
dispatched once, and hardens download handling for both local and remote
sessions.
- **Bug Fixes**
- Add a “handled” guard to dedupe dispatch across CDP progress, JS
fetch, and polling paths.
- Disable JS fetch fallback for local downloads (behind a
_use_js_fetch_for_local flag) to avoid double-dispatch.
- Improve remote handling: emit completion using suggestedFilename when
filePath is missing; never touch local FS.
- Move CDP handlers outside try, track handler tasks, and clean up on
completion for better stability.
<!-- End of auto-generated description by cubic. -->
This commit updates the DownloadsWatchdog class to specify exception types in the error handling logic, replacing broad exception catches with more precise ones (AssertionError, KeyError, AttributeError). Additionally, it adds a warning log when no downloads path is configured, ensuring better error visibility and handling during download operations.
This commit modifies the `download_file.py` example to replace the `BrowserSession` and `BrowserProfile` with the new `Browser` class. The changes streamline the code and enhance clarity by directly utilizing the `Browser` for managing downloads. The download path remains set to a temporary directory for testing purposes.
This commit updates the DownloadsWatchdog class to ensure that the downloads directory path is properly expanded and resolved to an absolute path. The changes include:
- Expanding the downloads path using `expanduser()` and `resolve()` methods to handle user home directories correctly.
- Updating the download behavior settings to use the expanded path.
- Ensuring that the downloads directory is created with the resolved path.
Additionally, the example feature for downloading files has been modified to use a temporary downloads path for testing purposes.
Auto-generated PR for branch: fix-disable-security-bug
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Fixes a bug where enabling disable_security overwrote --disable-features
and broke extensions. We now merge all --disable-features values,
de-duplicate them, and preserve order so both extension and security
flags are kept.
- **Bug Fixes**
- Merge all --disable-features args into a single flag; remove
duplicates while preserving order.
- Keep existing dedupe logic for other args unchanged.
- Add CI tests to ensure extension-related flags are preserved and
duplicate disable-features are consolidated.
<!-- End of auto-generated description by cubic. -->
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Adds an advanced example that integrates Playwright with Browser-Use
over a shared Chrome DevTools Protocol session. Lets Browser-Use
orchestrate tasks while Playwright handles precise form filling,
screenshots, and text extraction via custom actions.
- **New Features**
- Starts Chrome with remote debugging and connects both tools to the
same session via CDP.
- Adds Playwright-backed actions: playwright_fill_form,
playwright_screenshot, playwright_get_text.
- Demonstrates a flow on httpbin: fill form, take screenshot, extract
title, submit.
- Includes dependency checks, robust cleanup, and optional JPEG quality
for screenshots.
<!-- End of auto-generated description by cubic. -->
This commit updates the Playwright integration example in `playwright_integration.py` to provide a clearer overview of its key features. The previous detailed description has been condensed into a more concise format, highlighting the main functionalities of sharing a Chrome instance via CDP and the ability to perform actions with both Playwright and Browser-Use.
- **Documentation Update**
- Simplified the example description for better clarity and focus on key features.
This commit adds a new example demonstrating the integration of Playwright with Browser-Use, showcasing how to utilize Playwright for form filling, taking screenshots, and extracting text. The example is linked in the documentation and includes installation instructions and a comprehensive code sample.
- **New Documentation**
- Added a new Playwright integration example in `playwright-integration.mdx`.
- Updated `docs.json` to include the new example in the documentation index.
- Enhanced the `add.mdx` file with references to the Playwright integration example.
This enhances the usability of Browser-Use by providing clear guidance on leveraging Playwright's capabilities alongside it.
This commit introduces a new example demonstrating the integration of Playwright with Browser-Use. The example showcases how to start Chrome with CDP enabled, create custom actions for form filling, taking screenshots, and extracting text using Playwright's capabilities. It includes necessary dependency checks and a main function to orchestrate the workflow using an AI agent.
- **New Features**
- Implemented custom actions: `playwright_fill_form`, `playwright_screenshot`, and `playwright_get_text`.
- Added error handling and dependency checks for required libraries.
- Provided detailed documentation within the code for clarity on usage and functionality.
This patch ensures that if 'title' is present in the 'required' list, it
is correctly removed, thus restoring the functionality of the native
Gemini integration for structured output. This ensures that 'Response'
and 'Page summary' are two different objects.
<!-- This is an auto-generated description by cubic. -->
## Summary by cubic
Fixes invalid JSON schema for Gemini by removing 'title' from the
required list during schema cleanup. Prevents validation errors when
properties are replaced with a placeholder.
- **Bug Fixes**
- Update clean_schema to drop 'title' from required arrays, avoiding
"required references missing property" errors in
browser_use/llm/google/chat.py.
<!-- End of auto-generated description by cubic. -->
Auto-generated PR for: enable StorageStateWatchdog if user_data_dir
provided
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Enable StorageStateWatchdog when storage_state or user_data_dir is set,
so sessions with persistence configured save storage on shutdown with
safe defaults. Adds clear debug logs to show when it’s enabled.
- **New Features**
- Automatically attaches StorageStateWatchdog if storage_state or
user_data_dir is provided.
- Uses conservative defaults: auto_save_interval=60s and
save_on_change=false (saves on shutdown).
- Adds debug messages showing enable/disable status and which options
triggered it.
<!-- End of auto-generated description by cubic. -->
## Fix StorageStateWatchdog lifecycle management
### Problem
StorageStateWatchdog continues monitoring after browser closes, causing
CDP connection errors in logs.
### Solution
- Add `BrowserStopEvent` handler to cleanly stop monitoring when browser
closes
- Await `LoadStorageStateEvent` dispatch to ensure storage loads before
navigation
- Properly cancel monitoring task to prevent post-close errors
### Impact
- No impact while `StorageStateWatchdog` is commented out in
`attach_all_watchdogs()`
- No more error logs after browser session ends
- Ensures authentication/cookies are fully loaded before navigation
begins
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Stop StorageStateWatchdog when the browser closes and await storage load
on connect. This prevents CDP connection errors after shutdown and
ensures cookies/auth are loaded before navigation.
<!-- End of auto-generated description by cubic. -->
fix codec error while writing log, below is the detailed error message
```
--- Logging error ---
Traceback (most recent call last):
File "D:\IDE\Python\Python312\Lib\logging\__init__.py", line 1163, in emit
stream.write(msg + self.terminator)
UnicodeEncodeError: 'gbk' codec can't encode character '\U0001f4be' in position 63: illegal multibyte sequence
Call stack:
File "D:\WorkHome\git\gitee\azhengzz\browser-use\example_one.py", line 147, in <module>
asyncio.run(main())
File "D:\IDE\Python\Python312\Lib\asyncio\runners.py", line 194, in run
return runner.run(main)
File "D:\IDE\Python\Python312\Lib\asyncio\runners.py", line 118, in run
return self._loop.run_until_complete(task)
File "D:\IDE\Python\Python312\Lib\asyncio\base_events.py", line 673, in run_until_complete
self.run_forever()
File "D:\IDE\Python\Python312\Lib\asyncio\windows_events.py", line 322, in run_forever
super().run_forever()
File "D:\IDE\Python\Python312\Lib\asyncio\base_events.py", line 640, in run_forever
self._run_once()
File "D:\IDE\Python\Python312\Lib\asyncio\base_events.py", line 1992, in _run_once
handle._run()
File "D:\IDE\Python\Python312\Lib\asyncio\events.py", line 88, in _run
self._context.run(self._callback, *self._args)
File "D:\WorkHome\git\gitee\azhengzz\browser-use\example_one.py", line 131, in main
agent = Agent(
File "D:\WorkHome\git\gitee\azhengzz\browser-use\browser_use\utils.py", line 330, in wrapper
result = func(*args, **kwargs)
File "D:\WorkHome\git\gitee\azhengzz\browser-use\browser_use\agent\service.py", line 267, in __init__
self._set_file_system(file_system_path)
File "D:\WorkHome\git\gitee\azhengzz\browser-use\browser_use\agent\service.py", line 512, in _set_file_system
logger.debug(f'💾 File system path: {self.file_system_path}')
Message: '💾 File system path: C:\\Users\\Zheng\\AppData\\Local\\Temp\\browser_use_agent_068aff10-a1d9-7ee5-8000-803a64a9e569_1756360970'
Arguments: ()
```
<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Force UTF-8 encoding for log file handlers to fix UnicodeEncodeError
when writing non-ASCII characters (e.g., emojis) to log files,
especially on Windows with GBK locales. Logs now write reliably without
crashing.
- **Bug Fixes**
- Set encoding='utf-8' on debug and info FileHandlers to prevent 'gbk'
codec errors.
<!-- End of auto-generated description by cubic. -->
