MCP-Local-with-Nvidia-API-LLM

Cinder003/MCP-Local-with-Nvidia-API-LLM

3.2

If you are the rightful owner of MCP-Local-with-Nvidia-API-LLM and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.

FastMCP server for system orchestration with NVIDIA LLM integration.

Tools
2
Resources
0
Prompts
0

MCP-Local-with-Nvidia-API-LLM

FastMCP server for system orchestration: create/read files (TXT, CSV, XLSX, DOCX, JSON), manage folders, run shell commands, launch apps, list processes, and zip folders with safety guards; plus an NVIDIA LLM client with intelligent query classifier mapping natural language to tool calls, resilient fallbacks, parameter extraction, and file previews built-in.

Enhanced FastMCP System Documentation

Overview

The Enhanced FastMCP System is a comprehensive, AI-powered file and system orchestration platform that combines the power of NVIDIA's language models with FastMCP's robust server-client architecture. This system provides natural language interfaces for complex file operations, system management, and data processing tasks with intelligent query classification.

System Architecture

Core Components

  1. Enhanced FastMCP Server (complete_server.py)

    • Comprehensive system orchestration capabilities
    • Universal file creation and management
    • System operations and process management
    • Configurable security and access controls

  2. Ultra-Robust NVIDIA Client with Query Classifier (complete_client.py)

    • Intelligent Query Classification: Auto-routes queries as Knowledge, Action, or Hybrid
    • Natural language processing with NVIDIA AI models
    • Advanced intent classification with triple-fallback parsing
    • Multi-strategy command parsing with resilient error handling

  3. Configuration System

    • Server configuration (server_config.json)
    • Client configuration (client_config.json)
    • Environment variable support
    • Runtime parameter adjustment

Key Features

🧠 Intelligent Query Classification System

Query Categories
  • Knowledge Queries: Informational requests answered directly by LLM
    • Examples: "What is machine learning?", "Explain Python programming"
  • Action Queries: System operations executed via MCP tools
    • Examples: "Create file report.txt", "Run dir command", "Launch notepad"
  • Hybrid Queries: Combined knowledge + action responses
    • Examples: "Explain ML and create demo.py", "What is file compression and zip my folder"
Classification Workflow
# Enhanced client flow with classification
User Query → Query Classifier → Route Decision
                     ↓
    Knowledge ← [LLM Response] → Action → [MCP Execution]
                     ↓
                  Hybrid → [Both LLM + MCP]
Classification Methods
  1. Primary: LLM-based classification with prompt engineering
  2. Secondary: Pattern-based fallback classification
  3. Tertiary: Keyword scoring with confidence thresholds

🔧 Universal File Operations

File Creation Support
  • Text Files: .txt, .md, .log, .ini, .cfg, .conf
  • Programming Files: .py, .js, .html, .css, .json, .xml, .yaml, .sql
  • Office Documents: .xlsx, .xls, .docx, .doc, .rtf
  • Data Files: .csv, .tsv
  • Scripts: .bat, .sh, .ps1
Advanced File Management
# Examples of supported operations
create_file(path, content="", file_type="auto", working_directory=None)
create_folder(path, working_directory=None)
read_file(path, working_directory=None)
list_directory(path=".", working_directory=None)
copy_file(source, destination, working_directory=None)
move_file(source, destination, working_directory=None)
delete_file(path, working_directory=None)
search_files(pattern, directory=".", recursive=True)

🖥️ System Operations

Shell Command Execution
  • Cross-platform shell command support (Windows CMD, PowerShell, Linux Bash)
  • Configurable timeout settings
  • Output capture and formatting
  • Working directory context
Process Management
  • List running processes with filtering
  • Process monitoring and statistics
  • System resource information
  • Application launching capabilities
Archive Operations
  • ZIP file creation with folder structure preservation
  • Archive extraction with proper permissions
  • Batch compression operations
  • Integrity verification

🤖 Enhanced Natural Language Interface

Query Classification Features
# New client components
_setup_query_classifier()      # LLM-based query routing
_setup_knowledge_llm()         # Dedicated knowledge responses
classify_query()               # Primary classification method
knowledge_workflow()           # LLM-only workflow
action_workflow()              # MCP-only workflow
hybrid_workflow()              # Combined workflow
process_query()                # Main routing entry point
NVIDIA AI Integration
  • Supported Models: Meta Llama 3.1, Mistral, CodeLlama, and more
  • Advanced Parsing: Intent classification with confidence scoring
  • Context Awareness: Working directory and session state management
  • Fallback Strategies: Multiple parsing approaches for robustness
Enhanced Classification Patterns
{
  "query_classification": {
    "knowledge": {
      "keywords": ["what", "how", "why", "explain", "tell me", "define"],
      "confidence_threshold": 3
    },
    "action": {
      "keywords": ["create", "make", "run", "execute", "launch", "show"],
      "confidence_threshold": 3
    },
    "hybrid": {
      "patterns": ["explain X and create Y", "tell me about X and make Y"],
      "confidence_threshold": 5
    }
  }
}

🔒 Security & Configuration

Access Control
  • System-wide access control with configurable restrictions
  • Path validation and sanitization
  • Restricted directory protection
  • File size limitations
Enhanced Configuration
{
  "server": {
    "max_file_size": 104857600,
    "system_wide_access": true,
    "restricted_paths": ["C:\Windows\System32", "/bin", "/sbin"]
  },
  "nvidia": {
    "model": "meta/llama-3.1-8b-instruct",
    "temperature": 0.1,
    "max_completion_tokens": 1000
  },
  "classification": {
    "enable_query_classification": true,
    "fallback_to_action": true,
    "min_confidence_threshold": 3
  }
}

Usage Examples

Enhanced Natural Language Commands

Knowledge Queries (LLM Response Only)
User: "What is machine learning and how does it work?"
Classification: Knowledge
Response: Comprehensive LLM explanation about ML concepts, algorithms, applications

User: "Explain the difference between Python and JavaScript"
Classification: Knowledge
Response: Detailed comparison of both programming languages
Action Queries (MCP Execution Only)
User: "Create a text file called notes.txt with some project ideas"
Classification: Action
Response: File created via MCP server with specified content

User: "Show me all files in the Documents folder"
Classification: Action
Response: Directory listing via MCP list_directory tool
Hybrid Queries (Both LLM + MCP)
User: "Explain Python programming and create hello.py"
Classification: Hybrid
Response: 
🧠 Knowledge Response: [Detailed Python explanation]
⚡ Action Result: [hello.py file created successfully]

User: "Tell me about data analysis and create sample.csv"
Classification: Hybrid
Response:
🧠 Knowledge Response: [Data analysis concepts and methods]
⚡ Action Result: [sample.csv created with headers]

Enhanced Client Architecture Changes

New Query Processing Flow

Before (Action-Only Client)
User Query → Action Parser → MCP Tool → Result
After (Classification-Enhanced Client)
User Query → Query Classifier → Branch Decision
                    ↓
Knowledge: LLM Response → Direct Answer
Action: MCP Tools → Tool Execution Result  
Hybrid: Both Workflows → Combined Response

New Client Components

1. Query Classification System
async def classify_query(self, user_input: str) -> str:
    """Classify query into knowledge, action, or hybrid"""
    # LLM-based classification with fallback
    
def _fallback_classification(self, user_input: str) -> str:
    """Pattern-based classification backup"""
    # Keyword scoring and confidence thresholds
2. Workflow Routing
async def knowledge_workflow(self, user_input: str) -> str:
    """Handle pure knowledge queries with LLM"""
    
async def action_workflow(self, user_input: str) -> str:
    """Handle action queries with MCP tools"""
    
async def hybrid_workflow(self, user_input: str) -> str:
    """Handle queries needing both knowledge and actions"""
3. Enhanced Configuration
def _setup_query_classifier(self):
    """Setup LLM-based query classification"""
    
def _setup_knowledge_llm(self):
    """Setup dedicated knowledge response system"""

Classification Accuracy Metrics

Confidence Scoring
  • High Confidence (Score 5+): Direct routing
  • Medium Confidence (Score 3-4): Routing with validation
  • Low Confidence (Score <3): Fallback to pattern matching
Fallback Strategy
  1. Primary: NVIDIA LLM classification
  2. Secondary: Pattern-based keyword matching
  3. Tertiary: Default to action workflow

Technical Specifications

Enhanced Dependencies

Server Dependencies (Unchanged)
# Core Framework
fastmcp>=1.0.0
# [Previous dependencies remain the same]
Enhanced Client Dependencies
# NVIDIA AI Integration (Enhanced)
langchain-nvidia-ai-endpoints>=0.1.0
langchain-core>=0.2.0

# FastMCP Client
fastmcp>=1.0.0

# New: Enhanced natural language processing
# Additional utilities for classification
asyncio
re (for pattern matching)
typing (for enhanced type hints)

Performance Characteristics

  • Classification Time: < 200ms per query
  • Knowledge Response: 1-3 seconds (LLM dependent)
  • Action Execution: < 500ms for simple operations
  • Hybrid Operations: Combined timing of both workflows
  • Fallback Activation: < 50ms for pattern matching

Enhanced Error Handling

Four-Level Error Recovery
  1. Primary: NVIDIA AI query classification
  2. Secondary: NVIDIA AI action parsing
  3. Tertiary: Pattern-based intent classification
  4. Quaternary: Rule-based command parsing with user clarification

Advanced Configuration

Classification Tuning

# Adjust classification sensitivity
"classification_config": {
    "knowledge_keywords": ["what", "how", "why", "explain", "define", "tell me"],
    "action_keywords": ["create", "make", "run", "execute", "launch", "show"],
    "hybrid_patterns": ["explain.*and.*create", "tell.*about.*and.*make"],
    "confidence_thresholds": {
        "knowledge": 3,
        "action": 3, 
        "hybrid": 5
    }
}

Workflow Customization

# Custom workflow behaviors
"workflow_config": {
    "knowledge_workflow": {
        "max_response_length": 1000,
        "include_examples": true,
        "format_markdown": true
    },
    "action_workflow": {
        "confirm_destructive_actions": true,
        "show_execution_steps": true
    },
    "hybrid_workflow": {
        "knowledge_first": true,
        "combine_responses": true
    }
}

Version History

  • v1.0: Initial release with basic FastMCP integration
  • v1.1: Added NVIDIA AI natural language processing
  • v1.2: Enhanced configuration system and error handling
  • v1.3: Advanced intent classification and multi-strategy parsing
  • v1.4: Comprehensive file type support and system operations
  • v1.5: NEW - Intelligent Query Classification System
    • Added query classifier for Knowledge/Action/Hybrid routing
    • Implemented dedicated knowledge workflow with LLM responses
    • Enhanced hybrid workflow combining both LLM knowledge and MCP actions
    • Added classification confidence scoring and fallback mechanisms

This documentation covers the Enhanced FastMCP System with Intelligent Query Classification - a powerful, AI-driven platform that seamlessly combines natural language understanding with robust system operations through intelligent query routing.