klavis

Why we included this project

Most teams building AI agents hit the same wall early: wiring each tool to a model is fiddly, and once you expose a few dozen functions the context window fills up and the model starts picking wrong. Klavis treats that as an infrastructure problem rather than an agent design problem. It ships prebuilt Model Context Protocol servers for services like Gmail, Slack, or GitHub, with OAuth and API key handling already wired in, so you get authenticated tool access without writing credential plumbing yourself. The Strata layer on top exposes a single MCP server that surfaces tools by intent and trims output fields, which targets the token cost and choice overload that show up once you pass a few dozen tools. Self-host the servers via Docker, use the Python or TypeScript SDKs, or just talk to the platform over plain REST; that range suits anything from a notebook prototype to a production agent fleet. And because it is open source, you can read exactly how tool discovery and auth are handled before trusting it with your workloads.