Plan with Claude.
Execute Everywhere.

One Claude. Infinite execution. Zero token waste.

Get Started code View on GitHub

How It Works

Claude plans the strategy. AutoClaude dispatches tasks. Your chosen AI agent executes \u2014 all in the background, all automatic.

psychology

1. Claude Plans

Claude Code analyzes the problem, designs the architecture, and writes a detailed task file with step-by-step instructions.

send

2. AutoClaude Dispatches

With one MCP tool call, AutoClaude pipes the task to your chosen agent. Headless. Auto-approve. Instant.

terminal

3. Agent Executes

Your AI coding agent runs the task independently using its own token pool. Claude is free for the next planning session.

Choose Your Agent

AutoClaude supports any terminal-invocable AI coding CLI. Pick the tool that matches your subscription plan.

QW

Qwen Code

CLI

Alibaba / OpenRouter

GM

Gemini CLI

CLI Free Tier

Google 1M context

CX

Codex CLI

CLI OpenAI

OpenAI Apache 2.0

AI

Aider

CLI BYO Key

Multi-model Git native

OC

OpenCode

CLI 75+ Models

Privacy-first

CL

Cline CLI

CLI Agnostic

Model-agnostic

CS

Cursor AI

Clipboard

GUI Manual paste

add_circle

Custom

Your Tool

add_custom_agent

Don't see your tool? Use add_custom_agent in Claude to register any CLI tool.

Standardized Process

Every task produces a structured report. Clear role separation between Claude (planning) and agent (execution).

psychology
Strategy

Claude Code

Architecture design, task planning, code review, and final verification. Zero execution \u2014 planning only.

send
Bridge

AutoClaude

Validates tasks, dispatches to agents, captures output, generates standardized reports with cost tracking.

terminal
Execution

AI Agent

File operations, git commits, builds, deployments. Autonomous execution with auto-approve.

QWEN_GITHUB_SETUP_summary.md
# Task Report: QWEN_GITHUB_SETUP
 
## Role Separation
| Planner | Claude Code | Strategy, architecture, task authoring |
| Dispatcher | AutoClaude | Validation, dispatch, output capture |
| Executor | Qwen Code | git init, files, build, deploy |
 
## Completion Checklist
| Step | Role | Status |
| Architecture planning | Claude | \u2705 |
| Task file authoring | Claude | \u2705 |
| Task dispatch | AutoClaude | \u2705 |
| Git repo initialization | Qwen Code | \u2705 |
| README creation | Qwen Code | \u2705 |
| Landing page (index.html) | Qwen Code | \u2705 |
| GitHub Pages deploy | Qwen Code | \u2705 |
| Final verification | Claude | \u2705 |
 
## Completion Status
| Status | \u2705 Completed |
| Duration | 127s |
savings

Save 60-80% of Your Claude Tokens

Every task dispatched through AutoClaude is automatically tracked. Claude only spends tokens on planning \u2014 your agent handles the heavy execution.

~7K
Claude tokens per task (planning only)
~25K
Equivalent tokens saved (vs. all-Claude execution)
~$0.30
Cost saved per task (Opus 4.7 pricing)

Based on Claude Opus 4.7 API pricing (\$5.00/1M input, \$25.00/1M output). Actual savings vary by task complexity.

description

Project Report

Auto-generated after every task. Shows who did what, token usage, and cost savings.

# \ud83d\udcca AutoClaude Project Report
## \ud83d\udc65 Who Did What
| \ud83e\udde0 Claude Code | Strategy, planning, verification | ~7K tokens/task |
| \ud83d\udd17 AutoClaude | Dispatch, capture, cost tracking | 0 tokens (local) |
| \u26a1 Qwen Code | File ops, git, build, deploy | ~25K tokens/task |

## \ud83d\udcb0 Savings
| Total Tasks: 12 | Tokens Saved: ~216,000 | Cost Saved: $3.60 |

Call get_project_report in Claude to see your live report.

Version History

Every version planned by Claude, executed by an AI agent. 16 versions in 2 days.

v5.5 2026-05-09

verify_project + Task Quality Rules

6-in-1 health check tool. 7 task quality rules from 15-task failure analysis. Auto-clean orphan scripts.

v5.4 2026-05-09

First-Run Onboarding Wizard

check_agent tool — auto-detect installed CLI tools. CLAUDE.md guides new users through setup.

v5.3 2026-05-09

Project Report System

PROJECT_REPORT.md auto-generation + get_project_report MCP tool.

v5.2 2026-05-09

Terminal Output Beautification

Unicode box-drawing + emoji for all MCP tool responses.

View Full Changelog →

Features

Everything you need to multiply your coding throughput without multiplying your token costs.

visibility_off

Headless Background

No terminal windows. No pop-ups. Your agent runs silently in the background while you keep working.

bolt

Auto-Approve

Zero confirmation prompts. Your agent auto-approves all actions for fully autonomous execution.

hub

Multi-Agent

7 built-in agents. Switch freely between Qwen Code, Gemini CLI, Codex, Aider, and more.

content_copy

Clipboard Bridge

GUI tools like Cursor are supported too. Task content copied to clipboard automatically.

captive_portal

Windows Native

Toast notifications, speech synthesis, headless background spawn \u2014 built for Windows 10/11.

savings

Token Tracking

Automatic cost calculation per task. Cumulative savings report. Know exactly what you save.

format_quote

"The smartest way to code with AI isn't using one model for everything \u2014 it's using the right model for each job. Claude plans. Your agent executes. Everyone wins."

Getting Started

Three commands to connect Claude Code to any AI coding CLI.

# 1. Clone and install
git clone https://github.com/zhewenzhang/AutoClaude.git
cd AutoClaude && npm install && npm run build
# 2. Add to Claude Code settings
// In ~/.claude/settings.json:
"mcpServers": {
  "autoclaude": {
    "command": "node",
    "args": ["\AutoClaude\dist\index.js (or: command: autoclaude after npm i -g)"]
  }
}
# 3. Restart Claude Code and dispatch
Claude: Write a task file, then dispatch_task("TASK.md")