Skip to main content
MCP connects tools to AI models, but it does not run them — tai is the server that does. You bring a capability (a plain Python function, an agent, an external MCP server) and the runtime supplies the operational layer around it: loading everything from a manifest, access control, OAuth connectors to outside apps, background execution, monitoring, storage, and human-in-the-loop steps. Everything plugs in against a shared set of interfaces, so anyone can extend it, and each install is a self-contained system you host yourself.

Get started

Install the runtime and boot a working server in minutes.

Concepts

The core subsystems of the platform, one page per pillar.

Guides

Task-shaped how-to guides for using and extending the platform.

Marketplace

Browse the plugin index and install into a running server.

Integrations

Connect Claude Desktop, Cursor, and any MCP client.

Reference

Generated HTTP API, CLI, Python SDK, and ecosystem catalog.

Contributing

Where each repository’s contributing guide lives.