local-llm-mcp

richardbaxterseo/local-llm-mcp

3.1

If you are the rightful owner of local-llm-mcp 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.

The Local LLM MCP Server v2.1 is a Model Context Protocol server designed to facilitate task offloading to local LLMs running in LM Studio, with enhanced security and functionality.

Tools
9
Resources
0
Prompts
0

Local LLM MCP Server v3.0

License: MIT Node.js Version GitHub release npm version

An MCP (Model Context Protocol) server that enables Claude to offload tasks to local LLMs running in LM Studio. Version 3.0 introduces context-aware prompts and 5 powerful new tools for 90% token savings!

🚀 What's New in v3.0

Context-Aware Prompts

All major tools now support optional context for domain-specific analysis:

  • Project Types: WordPress, React, n8n, Node.js, and more
  • Framework-Specific: Best practices for each framework
  • Smart Suggestions: Context-aware refactoring and testing
  • 90% Token Savings: Offload routine tasks intelligently

5 New Powerful Tools

  1. generate_wordpress_plugin - Complete WordPress plugin generator
  2. analyze_n8n_workflow - n8n workflow optimization
  3. generate_responsive_component - Accessible component generator
  4. convert_to_typescript - JavaScript to TypeScript converter
  5. security_audit - Project-specific security auditor

Full Backward Compatibility

All existing tools work exactly as before - context is optional!

Documentation

  • 📚 - Full technical documentation
  • 🚀 - Step-by-step setup from zero
  • 📖 - All available tools
  • 🔄 - Upgrading to v3.0
  • 🔧 Troubleshooting - Common issues and solutions

Features

Core Capabilities

  • Code Analysis: Structure, patterns, complexity, security
  • Code Generation: Tests, documentation, refactoring suggestions
  • File Processing: Direct file analysis without manual reading
  • CSV Analysis: Filter and analyze data with custom criteria
  • TypeScript Conversion: Migrate JavaScript projects to TypeScript
  • Security Auditing: Project-specific vulnerability scanning
  • Component Generation: Create accessible, responsive components

Supported Project Types

  • WordPress (Plugins & Themes)
  • React (Apps & Components)
  • n8n (Nodes & Workflows)
  • Node.js (APIs & CLIs)
  • HTML/CSS Components
  • Generic JavaScript/TypeScript

Prerequisites

  1. LM Studio: Download and install from lmstudio.ai
  2. Node.js: Version 18 or higher
  3. A loaded model in LM Studio: Recommended models:
    • Qwen 2.5 Coder (best for code tasks)
    • DeepSeek Coder
    • CodeLlama variants
    • Any general model for non-code tasks

Installation

Method 1: NPM Global Install (Recommended)

npm install -g @richardbaxterseo/local-llm-mcp

Claude Config:

{
  "mcpServers": {
    "local-llm": {
      "command": "npx",
      "args": ["@richardbaxterseo/local-llm-mcp"],
      "env": {
        "LM_STUDIO_URL": "ws://localhost:1234",
        "LLM_MCP_ALLOWED_DIRS": "C:\\Users\\YourName\\Documents,C:\\Projects"
      }
    }
  }
}

Method 2: Local Installation

git clone https://github.com/richardbaxterseo/local-llm-mcp.git
cd local-llm-mcp
npm install
npm run build

Usage Examples

Basic Usage (Backward Compatible)

// Works exactly as before - no changes needed!
await local-llm:analyze_code_structure({
  filePath: "app.js"
});

Enhanced Context Usage (New in v3.0)

// WordPress Plugin Analysis
await local-llm:analyze_code_structure({
  filePath: "wp-content/plugins/my-plugin/my-plugin.php",
  context: {
    projectType: "wordpress-plugin",
    framework: "WordPress",
    frameworkVersion: "6.4",
    standards: ["WordPress Coding Standards", "PSR-4"]
  }
});

// React Component Testing
await local-llm:generate_unit_tests({
  filePath: "components/UserProfile.jsx",
  context: {
    projectType: "react-component",
    testFramework: "jest",
    dependencies: ["react-testing-library", "jest-dom"],
    coverageTarget: 80
  }
});

// n8n Node Analysis
await local-llm:analyze_n8n_workflow({
  workflow: {
    nodes: [...],
    connections: {...}
  }
});

New Tool Examples

Generate WordPress Plugin
await local-llm:generate_wordpress_plugin({
  pluginName: "Custom Analytics",
  description: "Track custom events in WordPress",
  features: ["admin dashboard", "REST API", "gutenberg block"],
  includeTests: true,
  phpVersion: "8.0"
});
Convert to TypeScript
await local-llm:convert_to_typescript({
  filePath: "src/utils/helpers.js",
  strictMode: true,
  preserveJSDoc: true,
  addTypeImports: true
});
Security Audit
await local-llm:security_audit({
  filePath: "api/auth.js",
  projectType: "node-api",
  checkTypes: ["injection", "authentication", "validation"],
  includeFixSuggestions: true
});

Token Savings Example

Without Context (Traditional Approach)

Claude reads 100 files → 100,000 tokens used
Claude analyzes each file → Complex context management
Result: Context window exhausted quickly

With Context (v3.0 Approach)

Local LLM analyzes 100 files → 0 Claude tokens used
Claude receives summaries → 5,000 tokens used
Result: 95% token savings, full project understanding maintained

Available Tools

Enhanced Tools (with optional context)

  • analyze_code_structure - Analyze code with framework awareness
  • generate_unit_tests - Create framework-specific tests
  • generate_documentation - Generate audience-appropriate docs
  • suggest_refactoring - Project-specific improvements

New Tools (v3.0)

  • generate_wordpress_plugin - Full WordPress plugin scaffolding
  • analyze_n8n_workflow - Optimize n8n workflows
  • generate_responsive_component - Create accessible components
  • convert_to_typescript - Migrate JS to TS intelligently
  • security_audit - Project-specific security analysis

Existing Tools

  • detect_patterns - Find design patterns and anti-patterns
  • validate_syntax - Check syntax and find bugs
  • suggest_variable_names - Improve naming conventions
  • analyze_file - General file analysis
  • analyze_csv_data - Filter and analyze CSV data
  • health_check - Verify LM Studio connection

Configuration

Environment Variables

  • LM_STUDIO_URL: WebSocket URL (default: ws://localhost:1234)
  • LLM_MCP_ALLOWED_DIRS: Comma-separated allowed directories
  • LLM_MAX_RETRIES: Max retry attempts (default: 3)
  • LLM_RETRY_DELAY: Retry delay in ms (default: 1000)

Security Settings

Configure allowed directories for file access:

# Windows
set LLM_MCP_ALLOWED_DIRS=C:\Projects,C:\Users\YourName\Documents

# Linux/Mac
export LLM_MCP_ALLOWED_DIRS=/home/user/projects,/home/user/documents

Best Practices

1. Use Context for Better Results

// ❌ Generic analysis
await local-llm:analyze_code_structure({ filePath: "plugin.php" });

// ✅ Context-aware analysis
await local-llm:analyze_code_structure({
  filePath: "plugin.php",
  context: { projectType: "wordpress-plugin" }
});

2. Offload Repetitive Tasks

  • File-by-file analysis
  • Boilerplate generation
  • Documentation creation
  • Basic refactoring

3. Keep Strategic Work in Claude

  • Architecture decisions
  • Complex integrations
  • Security-critical reviews
  • User experience design

Troubleshooting

LM Studio Connection Issues

# Test connection
npm test:connection

# Check if LM Studio is running on correct port
curl http://localhost:1234/v1/models

Common Issues

  1. "Server transport closed unexpectedly"

    • Ensure LM Studio is running
    • Check the URL matches your settings
    • Verify a model is loaded

  2. "Path not allowed"

    • Add directory to LLM_MCP_ALLOWED_DIRS
    • Use absolute paths

  3. "No response from LLM"

    • Check model is loaded in LM Studio
    • Verify model has enough context length
    • Try a different model

Contributing

See for development setup and guidelines.

License

MIT License - see for details.

Changelog

See for version history.

Links