Connects Claude or any MCP client to the Tuning Engines platform for fine-tuning code models on your repositories. Exposes 60+ tools that let your AI assistant create training jobs, estimate costs, monitor status, list models, and manage account balance through natural language. Ships two specialized agents out of the box: Cody for autocomplete fine-tuning using QLoRA, and SIERA for bug-fix specialists trained on synthetic error scenarios. Supports base models from 3B to 72B parameters across Qwen, CodeLlama, DeepSeek, and StarCoder families. Deliberately omits inference key creation and secret management for security, pushing those workflows to the CLI or web UI. Reach for this when you want to spin up domain-specific code models without infrastructure.
claude mcp add --transport stdio io.github.cerebrixos-tuning-engines uvx tuning-engines