akashkarnatak/claude-web — explained in plain English
Analysis updated 2026-05-18
Run Claude Code through a browser interface instead of the terminal, with formatted text, math, and syntax highlighted code.
Review and approve or deny tool permission requests from the assistant through an in-browser prompt.
Switch between using Claude Code in the terminal and in the browser while keeping the same conversation history.
Paste a screenshot or drag an image into a chat so the assistant can see and respond to it.
| akashkarnatak/claude-web | 0xradioac7iv/tempfs | 7vignesh/pgpulse | |
|---|---|---|---|
| Stars | 0 | 0 | 0 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 3/5 | 3/5 | 4/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Node 20.x and an Anthropic API key set in a local .env file, the server is meant to stay bound to localhost only.
claude-web is a local web application that lets a person run Claude Code, an AI coding assistant, through a browser interface instead of only in a terminal. It runs the Claude Code engine on a backend server and displays its output nicely in the browser, including formatted text, math notation, syntax highlighted code, and cards showing what tools the assistant used. The backend is written in Node and TypeScript, drives the Claude Code engine, and streams updates to the browser over a WebSocket connection that only accepts connections from the same computer by default. When the assistant wants to use a tool that needs permission, the browser shows an Allow or Deny prompt rather than asking in the terminal. Chats started in claude-web share the same history as chats started from the terminal, since the app reads and lists the same session files the command line version already keeps for a project. The interface tries to match what a person already gets in the terminal version of Claude Code, plus a few extras. It shows a status line with elapsed time, a live token count, and the current activity while the assistant works, and pressing Escape interrupts it. A keyboard shortcut cycles through permission modes such as accepting edits automatically or planning only. Typing a slash suggests available commands, and an at sign suggests project files. Every conversation lives at its own web address, so links can be shared or reopened after a refresh. When the assistant splits work across background subagents or runs a background shell command, a live panel shows each one with its status, elapsed time, and token count. If it asks the user a question through its question tool, the app renders real answer options and checkboxes instead of a plain approve or deny choice. There is also a search feature across all chats using regular expressions, and support for pasting or dragging images into a conversation so the assistant can see them. Setting it up means installing the Node dependencies, copying an example environment file, and adding an Anthropic API key, which stays only on the backend. It can run in a development mode with separate frontend and backend processes, or as a single built production app. The project requires Node 20.x and is built to run locally rather than be exposed to a network. The README does not mention a software license.
claude-web is a local browser interface for the Claude Code AI coding assistant, mirroring the terminal experience with rich formatting, tool cards, and shared chat history.
Mainly TypeScript. The stack also includes TypeScript, Node.js, React.
No license is stated in the README, so usage rights are unclear.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.