MCP 服务器

Manus Credit Optimizer

An MCP server that reduces Manus AI credit usage by up to 75% through intelligent prompt compression, smart model routing, and intent classification. It provides tools to analyze and optimize prompts for maximum efficiency without sacrificing quality.

README

🚀 Credit Optimizer v5 for Manus AI

Cut your Manus AI credit usage by 30-75% with ZERO quality loss.

Audited across 53 real-world scenarios. Works as an MCP Server (Claude, Cursor, Copilot, Codex, Windsurf, Cline) or as a native Manus Skill.

📊 Results

Metric	Before	After	Improvement
Average credits per task	100%	53%	47% reduction
Quality score	9.2/10	9.2/10	0% loss
Tasks audited	-	200+	Verified
Scenarios tested	-	53	All passing

🎯 What It Does

Credit Optimizer analyzes your prompts and tasks before execution, then applies intelligent optimization strategies:

Smart Model Routing — Routes simple tasks to Standard mode, complex tasks to Max mode
Intent Classification — Detects 12 task categories (code_fix, creative_writing, data_analysis, etc.)
Prompt Compression — Removes redundancy while preserving all semantic meaning
Batch Detection — Identifies tasks that can be parallelized for fewer credits
Context Hygiene — Flags unnecessary context that inflates token usage
Output Format Optimization — Suggests efficient output formats (file vs inline)

⚡ Quick Start

As MCP Server (Claude Desktop, Cursor, etc.)

# Install
pip install fastmcp

# Clone this repo
git clone https://github.com/rafsilva85/manus-credit-optimizer.git
cd manus-credit-optimizer

# Run the server
python -m mcp_credit_optimizer

Add to your MCP client config (e.g., Claude Desktop claude_desktop_config.json):

{
  "mcpServers": {
    "credit-optimizer": {
      "command": "python",
      "args": ["-m", "mcp_credit_optimizer"],
      "cwd": "/path/to/manus-credit-optimizer"
    }
  }
}

As Manus Skill (Native Integration)

Purchase the full Manus Skill at creditopt.ai
Install as a Manus Skill following the included instructions
The skill auto-activates on every task — no manual intervention needed

🔧 MCP Tools Available

Tool	Description
`analyze_prompt`	Analyze a prompt and get optimization recommendations with estimated savings
`optimize_prompt`	Get an optimized version of your prompt ready to use
`estimate_savings`	Quick estimate of potential credit savings for a task description

📋 Example Usage

You: analyze_prompt("Build me a React dashboard with charts, authentication, and a database backend")

Credit Optimizer: 
  ✅ Intent: code_generation (complex)
  ✅ Recommended: Max mode (complex multi-component task)
  ✅ Optimization: Split into 3 sequential tasks
     1. Database schema + API routes
     2. Authentication flow  
     3. React dashboard + charts
  ✅ Estimated savings: 35-45%
  ✅ Quality impact: None

🏗️ Architecture

Prompt Input → Intent Classifier → Complexity Scorer → Strategy Selector
                                                            ↓
                                              ┌─────────────────────────┐
                                              │ • Model Router          │
                                              │ • Prompt Compressor     │
                                              │ • Batch Detector        │
                                              │ • Context Hygiene       │
                                              │ • Output Optimizer      │
                                              └─────────────────────────┘
                                                            ↓
                                              Optimized Recommendations