Aito MCP server

Aito Python SDK

:Info: Check Github for the latest source code and documentation for more information. :Maintainer: Aito

About

The Aito Python SDK is an open-source library that helps you to integrate your Python application to Aito _ quicker and more efficiently.

The SDK also includes the Aito Command Line Interface (CLI) _ that enables you to interact with Aito using commands in your command-line shell, e.g: infer a table schema from a file or upload a file to Aito.

API versions

Each Aito API version has its own package, whose meaning never changes:

.. code-block:: python

from aito.v2 import Client # the v2 API (from aito.v1 import Client for v1)

# the public read-only sandbox; use your own instance URL and key in production client = Client('https://shared.aito.ai/db/aito-demo', 'yg4rTlXkqDzm4y8gPeY75HCKaNwfbTQ2si64ONTi', env='v2') prediction = client.predict( from_table='invoices', where={'Description': 'cloud services'}, predict='GLCode') print(prediction.first.value, prediction.first.probability) # E002 0.83...

aito.Client is the default version, and moves only on a major release of this package: since aitoai 1.0 it is the v2 client (0.x: v1). Quickstarts can use it; production code should import the version it means, aito.v1 or aito.v2, so an upgrade cannot change the API underneath it. To stay on v1: from aito.v1 import Client, or pin aitoai<1.

The pre-0.7 paths (aito.client, aito.client.v2, aito.api) were removed in 1.0; importing one raises an ImportError naming its replacement. The command-line tool still talks to the v1 API in 1.x: v2 support arrives as an opt-in in a 1.x minor release, and the CLI's default moves to v2 only in 2.0. The design is written up in docs/versioned-namespaces.md.

pip install aitoai installs the API clients. The command-line tool, schema inference and file conversion need the dataframe toolchain: pip install 'aitoai[cli]'.

Run Aito locally

With Docker installed, aito start runs a local Aito and stores its keys, so the client needs no arguments:

.. code-block:: console

$ pip install aitoai $ aito start # pulls a pinned image, listens on http://127.0.0.1:9005, prints the keys $ python -c "import aito; print(aito.Client().get_schema())"

aito stop stops it (data and keys are kept), aito status checks it, aito keys prints the keys again (--rotate replaces them), aito logs shows its log and aito upgrade moves to the engine version this package pins. These work without the [cli] extra. From Python, aito.Client() with no arguments uses AITOURL and AITOAPI_KEY if set, else the local server's profile.

examples/v2_quickstart.py is a runnable end-to-end example that creates a collection, loads it, predicts, explains, evaluates and drops it again.

Use Aito from an AI assistant (MCP)

pip install 'aitoai[mcp]' (Python 3.10+) adds aito-mcp, an MCP server for coding agents and assistants. Its tools are predict, recommend, relate, match, search, query, evaluate and schema/upload. Each tool's description says when Aito fits and when another tool (a language model, a trained model, a search engine) is the better choice. Results carry $p and a flattened $why. The server finds the instance the same way as aito.Client(). It is read-only unless started with AITOMCPALLOW_WRITES=1.

.. code-block:: console

$ claude mcp add aito -- aito-mcp # Claude Code; any MCP client runs aito-mcp over stdio

Shortened. The full README is on GitHub.