AMD Ryzen AI MCP Server

AMD Ryzen AI Model Context Protocol (MCP) Server for Cursor IDE.

This MCP server provides semantic search and access to AMD Ryzen AI documentation and code examples from the amd/RyzenAI-SW repository and AMD Ryzen AI documentation.

Setup

Prerequisites

• Python 3.8 or higher
• pip package manager

Installation

1. Clone the repository:

   git clone https://github.com/amd/ryzenai-mcp-server
   cd ryzenai-mcp-server

2. Install dependencies:

   pip install -r requirements.txt

3. Configure the server:

   • The server will work out of the box
   • Optionally set GITHUB_TOKEN environment variable for higher GitHub API rate limits
   • Get a token from GitHub Settings → Developer settings → Personal access tokens

Configuration

The server supports the following environment variables:

• GITHUB_TOKEN (optional): GitHub personal access token for higher API rate limits. Requires public_repo scope.
• RYZENAI_INDEX_REFRESH_HOURS (optional): Hours between automatic index refreshes. Default 24. Set to 0 to disable auto-refresh.
• RYZENAI_INDEX_MAX_FILES (optional): Max files to download per refresh (used by auto-refresh and by the refresh tool). Default 100.

Features

• ✅ Semantic Search: AI-powered semantic search of AMD Ryzen AI repository using embeddings
• ✅ Enhanced Code Chunking: Tree-sitter AST-based chunking for better code understanding
• ✅ Documentation Access: Read AMD Ryzen AI documentation pages directly
• ✅ Repository Browsing: List directories, find files, and read code from the repository
• ✅ Smart Indexing: Automatic download and indexing of repository files with embeddings

Available Tools

The MCP server provides 9 tools:

1. read_ryzenai - Read AMD Ryzen AI documentation pages
2. search_ryzenai_sw_keyword - Keyword search in the Ryzen AI repository
3. search_ryzenai_sw_enhanced - Enhanced semantic search with AI embeddings
4. read_ryzenai_sw - Read specific files from the repository
5. list_ryzenai_sw_directory - List contents of a directory
6. find_ryzenai_sw_files - Find files by name pattern or type
7. download_and_index_ryzenai_enhanced - Download and index repository with AI embeddings
8. get_enhanced_embedding_status - Check embedding index status
9. refresh_ryzenai_index - Clear the index and re-download/re-index (use after repo or docs updates)

Refreshing the index

The index is refreshed automatically every 24 hours (configurable via RYZENAI_INDEX_REFRESH_HOURS). A background thread clears the embedding store and re-downloads/re-indexes the repository so you get the latest Ryzen AI repo and docs without doing anything.

• Default: refresh every 24 hours (reasonable for tracking upstream releases).
• Disable: set RYZENAI_INDEX_REFRESH_HOURS=0.
• Manual refresh: you can still ask the AI to run refresh_ryzenai_index anytime (e.g. "Refresh the Ryzen AI MCP index"); that also resets the 24h timer.
• Status: use get_enhanced_embedding_status to see last_refresh_iso and auto_refresh_hours.

Usage Examples

Once installed in Cursor, you can ask:

• "Find inference examples for Llama models"
• "Read the getting started guide from Ryzen AI docs"
• "Write the python code to compile and run resnet50 model on AMD RyzenAI NPU"
• "Write the python code to compile and run the distilbert-base-uncased-finetuned-sst-2-english model (the fine-tuned checkpoint of DistilBERT-base-uncased, trained on the SST-2 dataset)"

Requirements

See requirements.txt for full dependency list. Key dependencies include:

• mcp>=1.14.0 - Model Context Protocol
• sentence-transformers>=2.2.0 - For AI embeddings
• chromadb>=0.4.0 - Vector database
• langchain-community>=0.0.20 - For text chunking
• tree-sitter-languages>=1.10.0 - For AST-based code chunking

Documentation

For detailed setup instructions, see the Setup Guide.

Repository

• GitHub: https://github.com/amd/ryzenai-mcp-server
• AMD Ryzen AI Docs: https://ryzenai.docs.amd.com
• RyzenAI-SW Repository: https://github.com/amd/RyzenAI-SW

License

MIT License