A new project on GitHub, CCBaga, packages local maintenance, browser checks and system region settings into a single desktop application. According to the project's README, the toolkit runs on macOS, Linux and Windows and aims to give users a cleaner workspace and an environment they can actually verify. The project page was published on 13 October 2026, and it is available under the liseami account on GitHub.
The tool is designed for people who work with Claude Code and Claude Desktop on their own machines and who also rely on a companion browser for web tasks. Instead of spreading these jobs across several utilities, CCBaga tries to handle them from one place.
Two applications, one workflow
The README describes two cooperating applications. CCBaga-A handles the native operations on the host computer. CCBaga-B runs an isolated Fingerprint Chromium profile in which the browser is used. Measurements taken in the browser are sent back to the host through a local session, so the main application can compare what the browser reports with the machine and the network exit.
A central idea in the documentation is the separation between changing a setting and confirming its effect. The project treats these as two distinct steps, so a user who rotates a browser fingerprint or changes a regional setting is expected to run a check afterwards rather than assume the change worked.
For everyday use, the README highlights an Easy setup page. It places browser cache cleanup, fingerprint rotation, timezone and language reset, and identity archiving on one screen, with a region selector at the top. More advanced controls sit in the sidebar under Network, Claude identities, Browser and Computer. The interface is available in English and Chinese, but changing the interface language does not alter the language used by the computer or the browser.
Managing Claude identities
The most detailed section of the README concerns Claude identities. The feature manages the local state of both Claude Code and Claude Desktop. A user can archive the current state and start from an empty identity, or restore a previously saved identity. Each switch saves the outgoing identity first, so the previous state is not lost.
Users can also choose what stays in the workspace during a switch. According to the documentation, the retention options cover:
- skills and commands
- plugins and MCP configuration
- personal settings
- project records and conversation history
- caches and logs
Login data, by contrast, always travels with an archived identity. The README notes that project source code outside Claude's own data directories is not scanned, and that a confirmation dialog summarises the changes before recognised Claude processes are closed.
The archive itself is described as an authenticated, encrypted container stored outside Claude's working directories. The project uses a local SQLite database to track names, policies and recovery state. If a switch fails, the app rolls back, and an interrupted switch can be recovered. The README is candid about the limits of this protection: ordinary archive tools cannot open the container, but file ownership or administrator access is not an absolute secrecy boundary.
Some limitations are also spelled out. Expired or revoked server-side logins may require signing in again. The app manages the default local directories only, not every custom Claude profile, and custom CLAUDE_CONFIG_DIR locations are not automatically included. The documentation says the app reports this rather than silently claiming such locations were switched.
Browser checks and optional network setup
The browser verification feature measures the actual environment the browser presents. The README lists the checks as covering the network exit, country, timezone, language, platform, CPU count, engine version and location settings. A success page appears only when every measured check matches and no measurement error remains.
The project is careful about what a green result means. The README states that DNS and WebRTC leak checks have not been independently verified, and that a passing result describes the measured checks only. It is not a guarantee of anonymity or of acceptance by any online account service.
Network setup is a separate and optional component. It uses a Network Skill that follows the Agent Skills specification and is installed separately from the applications. The documentation says it supports authorised Debian 12 and 13 servers with root key login on port 22, and that it sets up Xray and Reality configuration after server self-checks pass. The README also notes that the subscription URL uses HTTP and contains credentials, and recommends preferring the file retrieved over SSH. Provider links in the documentation point to DMIT and Webshare, but the README stresses that local cleanup does not require a VPS or a proxy subscription.
Platform support and installation
The README sets clear platform boundaries. macOS 12 or later is supported on Apple Silicon. Linux is supported on x86_64 with GTK 3 and WebKitGTK 4.1, with packages in DEB, RPM, AppImage and Arch formats. Windows 10 and 11 on x86_64 are supported through a single NSIS installer. Intel Macs and Linux ARM64 systems are not supported.
The Windows installer is not Authenticode signed, the documentation warns, and it may trigger Microsoft SmartScreen. Managed security policies can block it altogether. On macOS, the README says a single DMG contains both applications, and users are asked to drag them into the Applications folder before launching CCBaga-A, which starts the companion browser automatically.
Updates are checked quietly at startup, but installation begins only when the user chooses to update and restart. Downloads are verified before installation, and the project describes how failed downloads can be retried.
Privacy, permissions and what remains unverified
The README makes several privacy commitments. Identities are stored locally, secrets are kept out of the interface and logs, and the apps have no SSH key picker or server deployment command. Network checks do make real requests: exit detection uses ipwho.is, with ipapi.co as a fallback on the same route, and the project notes that IP location is only a service estimate.
Some features require system permissions. Timezone changes need macOS administrator authorisation, Keychain access follows standard system prompts, and location controls may require Accessibility permission or a manual step in System Settings.
The documentation also lists what its automated tests do not cover. Those tests use disposable fixtures and local test servers, so they do not deploy a real VPS, delete real Claude account data or perform administrator-authorised system changes. Real SSH connectivity, client-side TUN routing and permission-dependent operations must be checked in the target environment.
CCBaga's original code is released under the MIT License, while third-party components keep their own terms. The project credits Tauri, React and Fingerprint Chromium among its building blocks.
For users who manage several Claude setups or who test browser configurations on their own machines, CCBaga offers an organised alternative to manual file handling. Its strongest feature is arguably its insistence on verification, but the README's own caveats make clear that users should read the limitations before relying on any result.
Related: Reduce Claude Code Token Costs With optimAIzr CLI · Anthropic's OSS Scanner Automates Security Checks for Open Source
Comments (0)