Roadmap

Public roadmap for gpu-mcp-server. Updated quarterly.

Q3 2026 (Current)

• [x] Core MCP server with stdio transport
• [x] list_gpus, get_gpu_metrics, gpu_summary tools
• [x] get_gpu_processes tool
• [x] NVML-based collector with MIG support
• [x] Mock collector for testing without GPU hardware
• [x] GitHub Actions CI
• [x] Publish to MCP servers registry
• [x] First tagged release (v0.1.0)
• [x] Container image on GitHub Container Registry (ghcr.io)

Q4 2026

• [ ] Streamable HTTP transport (run as a network service)
• [ ] MCP resources expose GPU info as context for agents
• [ ] Per-process GPU usage (PID-level memory and compute attribution)
• [ ] Prometheus metrics endpoint (optional, alongside MCP)
• [ ] Integration tests with Claude Desktop and Goose

Q1 2027

• [ ] Multi-node support — aggregate metrics from remote hosts
• [ ] GPU event notifications (thermal throttling, ECC errors, XID events)
• [ ] OpenSSF Best Practices badge
• [x] Helm chart for Kubernetes deployment

Q2 2027

• [ ] AMD ROCm support (via rocm-smi)
• [ ] Intel GPU support (via oneAPI Level Zero)
• [ ] MCP sampling integration — let agents request GPU snapshots at intervals
• [ ] AAIF project proposal submission (Growth stage)

How to contribute

Pick an item from the roadmap, open an issue to discuss the approach, and submit a PR. See Contributing for details.