WebCodex connects a chat-based AI assistant to the machine where your code already lives. You can ask the assistant to inspect a repository, modify code, run tests, use Git, or investigate a failure, and the repository stays where it is rather than being copied into a hosted workspace.
An AI client reaches WebCodex over the Model Context Protocol (MCP) or HTTPS, and WebCodex in turn reaches the repository, Git checkout, compilers, tests, and other developer tools on the host. Two setups are documented: a Server plus Runner installation for durable access to several projects, and webcodex share, run inside a single repository, which starts a temporary single-project restricted environment and prints connection values that stop working when the command exits.
Public HTTPS, a Cloudflare Tunnel, and the OpenAI Secure MCP Tunnel are ways for a client to reach the Server rather than separate modes. Platform coverage is Linux x64 and arm64 for share, Server, and Runner workflows; macOS x64 and arm64 for share and Runner; and Windows x64 and arm64 for the CLI, Runner, a local foreground Server, and explicit sharing, where WebCodex-managed Windows Server services are unsupported.
Features
- Code understanding and editing: read, search, inspect, and make guarded changes inside configured projects
- Real toolchain: run commands, tests, formatters, compilers, and project-specific tooling on the machine that owns the repository
- Git work: inspect status and diffs while keeping repository operations visible and reviewable
- Long-running work: jobs stay observable instead of requiring one model turn to stay open indefinitely
- Human review: a Runtime Console and task workflow to guide, cancel, accept, or reject work where those actions are available
- Temporary sharing: webcodex share starts a short-lived, single-project environment with a credential that expires when the command exits
- Reach options: public HTTPS, Cloudflare Tunnel, or the OpenAI Secure MCP Tunnel connect a client to the Server
- Scoped access: only the project roots you register are exposed to the assistant, which can read files, modify them, and execute commands within those boundaries
