TabPFN MCP, gives LLMs tools for predictions on tabular data (beta)

TabPFN MCP is an MCP server that exposes TabPFN-based tabular prediction as tools to any MCP-enabled LLM. Within a conversation, the LLM can call these tools to train and run classification or regression on tabular data you provide, without you having to build or tune traditional ML pipelines.

TabPFN MCP is designed for supervised learning on tabular data, mainly classification and regression. Typical tasks include churn prediction, lead scoring, risk flagging, outcome modeling for experiments, and forecasting based on structured metrics.

No. TabPFN is a pretrained model that works out of the box, so you typically do not need to design architectures or tune hyperparameters. Your main responsibility is to prepare sensible tabular inputs and let the LLM orchestrate calls to the MCP tools.

TabPFN MCP runs wherever you deploy the MCP server, so your tabular data can stay within your own infrastructure or environment. The MCP layer separates data access from model logic, letting you control how data is loaded, filtered, and secured.

Pricing is currently not publicly specified, and the project is in beta. It is well-suited for experimentation, prototyping, and internal tools; you should evaluate performance, reliability, and licensing before using it in strict production environments.