Skip to content

ADLX MCP Servers

Navigation: Programming with ADLX

In this section

The Model Context Protocol (MCP) is an open-source standard to connect AI models directly to external data sources and tools. To explore the step-by-step implementation tutorials, search online for the “Model Context Protocol developer documentation.” The ADLX MCP package provides MCP servers and clients that expose AMD ADLX driver features to LLM-based agents, enabling AMD GPUs to be managed through natural language.

Each MCP server wraps a set of ADLX interfaces as callable tools. An MCP client interprets a user’s natural-language request, selects the appropriate tool, invokes it on the server, and returns a response. Internally, the servers call ADLX through the ADLX Python binding.

Prerequisites

  • Windows 10 (64-bit) or later
  • Python 3.12 or later
  • The uv package manager (to explore the complete command reference, installation guides, and project workflows for uv, search online for the “official Astral uv documentation”)
  • An AMD GPU with ADLX-compatible drivers installed
  • The ADLX Python wheel (built from ADLXPybind)

Architecture

User Query --> MCP Client --> Tool Selection --> MCP Tool Call --> ADLX (Python binding) --> AMD Driver

Each server can run over stdio (default, for subprocess communication) or streamable HTTP (for network access).

  • MCP Servers describes the available servers and the AMD driver features they expose.
  • MCP Clients describes the natural-language client interface.

Subpages

See Also: Building Python bindings for ADLX, MCP Clients, MCP Servers