Keryx

Keryx is an AI developer tool — a full-stack TypeScript framework that lets you write one action class and deploy it automatically across five transports: HTTP, WebSocket, CLI, background tasks, and MCP (Model Context Protocol) for AI agents. Built on Bun for native TypeScript with no compilation step, its core idea is a unified action system: a single action class becomes an API endpoint, a WebSocket handler, a CLI tool, a background task runner, and an MCP server an AI agent can call. Concrete features from the site include Zod-based type-safe validation that auto-generates OpenAPI docs and CLI help, real-time PubSub channels over Redis with middleware authorization, built-in background workers with fan-out patterns, Drizzle ORM with auto-migrations, OpenTelemetry metrics with structured logging, and AI-native plumbing such as OAuth 2.1 and per-session MCP servers. It targets TypeScript teams building APIs that need multi-transport access and developers wiring AI agents in over MCP without maintaining a separate server. Keryx is released under the MIT License, so it is open source. The trade-off is that it is Bun-first and TypeScript-only, so it suits teams already committed to that stack rather than polyglot backends.

Building multi-transport TypeScript APIs (HTTP, WebSocket, CLI, tasks, MCP) from a single action class, for TypeScript teams and AI-agent integrators.