WebCodex connects cloud AI assistants such as ChatGPT and Claude to repositories and developer tools on your own machines. Instead of copying a project into a hosted workspace, the assistant reaches the machine that already holds the repository over MCP (Model Context Protocol) and HTTPS, and works with its files, Git checkout, compilers, tests, and project tooling there.
Through WebCodex an assistant can read, search, and edit code inside configured project roots, run commands, tests, and formatters, and inspect Git status and diffs. Long-running work runs as observable jobs, so a task does not depend on a single model turn staying open, and a Runtime Console, workflow session evidence, and diff review support human review of what the agent did.
The system is split into a Server and one or more Runners, with a Desktop app and a CLI on top. One computer can host everything, or a self-hosted Server can coordinate Runners on several machines that each own repositories. A temporary share mode exposes a single repository for a short trial. Unified installers for Windows, macOS, and Debian or Ubuntu are still being validated before release.
Features
- Code access: reads, searches, and edits files inside configured project boundaries
- Real toolchain: runs commands, tests, formatters, compilers, and project-specific tooling on the machine that owns the repository
- Git: inspects status and diffs, keeping repository operations visible and reviewable
- Long-running jobs: execution stays observable beyond a single model turn
- Human review: Runtime Console, workflow session evidence, job history, and diff review
- Multi-machine setup: a self-hosted Server coordinates Runners on each computer that holds repositories
- Quick trial: a temporary share mode connects one repository for a short session
- Clients: works with ChatGPT, Claude, and other MCP clients
