Aito

Lets an assistant run predictions, recommendations, matching and search over your own Aito predictive database, with probabilities and evidence.

What it does

An MCP server for Aito, a predictive database, exposing tools named predict, recommend, relate, match, search, query, evaluate, get_schema, put_schema and upload_rows over the Aito v2 API. Results include $p, a probability you can threshold, and $why, the supporting evidence. Tool descriptions state when Aito fits and when a language model, a trained model or a search engine is the better choice. It is read-only unless writes are explicitly enabled.

When to use it

Use it when an assistant should answer questions or make recommendations directly from data already stored in an Aito instance, without training a model. It suits teams that already run Aito and want predictive queries, matching or schema inspection inside a coding agent or chat assistant.

Requirements

Runs locally over stdio, typically via uvx aito-mcp, so a Python runtime with uv is needed. It requires an Aito instance, located either by the AITO_URL and AITO_API_KEY environment variables or by the profile written by the aito start command. A read-only API key is sufficient unless writes are enabled.