ClaudesClaude

Problem

You're building a local web app that needs LLM API access. But paying per-token through the Anthropic API adds up fast — especially for hobby projects, prototyping, or personal tools where you're iterating constantly. You're already paying for a Claude Pro/Max subscription, but there's no way to use it programmatically.

Solution

ClaudesClaude is a local API proxy that mirrors the Anthropic Messages API — but routes requests through your existing Claude subscription instead of the paid API. Your code talks to localhost:3456 using the same Anthropic SDK and request format it already uses. Just swap the baseURL and you're done. No API keys, no per-token billing.

Two modes:

| Mode | Command | Speed | Requires | |---|---|---|---| | CLI tunnel (default) | npx claudesclaude | ~3-5s per request | Claude Code CLI installed | | Browser automation | npx claudesclaude --browser | ~15-30s per request | Google Chrome |

Quick Start

Default (CLI tunnel)

Pipes requests through the Claude Code CLI. No browser needed, instant startup.

npx claudesclaude

Requires claude CLI in your PATH. Install it from claude.ai/claude-code.

Browser Mode

Automates Claude's web UI with Puppeteer. Slower but works without the Claude Code CLI.

npx claudesclaude --browser

Opens Chrome, navigates to claude.ai, waits for you to log in (first time only — session is persisted).

Usage

With the Anthropic SDK

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "http://localhost:3456",
  apiKey: "unused",
});

const response = await client.messages.create({
  model: "claude-sonnet-4-20250514",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello!" }],
});

console.log(response.content[0].text);

With curl

curl -X POST http://localhost:3456/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-20250514",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Streaming

curl -N -X POST http://localhost:3456/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-20250514",
    "max_tokens": 256,
    "stream": true,
    "messages": [{"role": "user", "content": "Tell me a story."}]
  }'

API Endpoints

| Endpoint | Method | Description | |---|---|---| | /v1/messages | POST | Send a message (matches Anthropic API format) | | /v1/models | GET | List available models | | /health | GET | Check if the session is ready | | /debug/screenshot | GET | PNG screenshot of browser (browser mode only) | | /debug/info | GET | Current page URL and title (browser mode only) |

Environment Variables

| Variable | Default | Description | |---|---|---| | PORT | 3456 | API server port | | HEADLESS | false | Run Chrome headless (browser mode only) | | CHROME_PATH | auto-detected | Path to Chrome/Chromium (browser mode only) | | CLAUDE_CLI_PATH | claude | Path to Claude Code CLI (CLI mode only) |

How It Works

CLI Tunnel Mode

1. Receives an API request matching the Anthropic Messages format
2. Spawns claude -p --output-format json as a subprocess
3. Passes the prompt via stdin, reads structured JSON from stdout
4. Returns the response in Anthropic API format
5. For streaming: uses --output-format stream-json --include-partial-messages

Browser Mode

1. Launches your system Chrome via puppeteer-core
2. Navigates to claude.ai and waits for login
3. Persists session in ~/.claudesclaude/browser-data/
4. For each request: opens a new conversation, pastes the message, waits for response
5. Extracts response via network interception or DOM parsing
6. Returns in Anthropic API format

Limitations

• CLI mode: Requires Claude Code CLI with an active subscription. Each request spawns a new process.
• Browser mode: Requests processed one at a time (~15-30s each). System prompts prepended as text. No image support. Claude.ai UI changes may break selectors.
• Both modes: Token counts may be estimates. Model selection maps to Claude CLI aliases (sonnet/opus/haiku).

Disclaimer: This tool may violate Anthropic's Terms of Service. Use at your own risk.

License

MIT