Skrun

Skrun is an open-source CLI tool that turns any Agent Skill (SKILL.md) into a callable REST API via POST /run. It supports multiple AI model providers, maintains stateful agent memory across runs, and integrates with the MCP (Model Context Protocol) ecosystem. Built in TypeScript and licensed under MIT, Skrun is designed for developers who want to deploy AI agents as production-ready APIs with minimal friction.

• Agent Skill Import — Use skrun init --from-skill ./path-to-skill to read an existing SKILL.md and auto-generate an agent.yaml config with just 2-3 prompts.
• POST /run API — Every agent is exposed as a typed REST endpoint with structured inputs and outputs, making agents consumable by any HTTP client.
• Multi-model Support — Configure Anthropic, OpenAI, Google, Mistral, or Groq as the model backend in agent.yaml, with automatic fallback between providers.
• Stateful Agents — Agents persist key-value state across runs, enabling memory and comparison between executions (e.g., SEO audit that remembers previous results).
• Tool Calling — Supports two approaches: local CLI scripts bundled with the agent (scripts/) and MCP servers via npx, compatible with the standard MCP ecosystem.
• Local Dev Server — Run skrun dev to start a local server at http://localhost:3000 with live reload and POST /run ready for testing.
• Built-in Testing — Define test cases in agent.yaml and run skrun test to validate agent outputs with scored assertions.
• Registry & Deployment — Use skrun build, skrun push, and skrun deploy to package and publish agents to a local or remote registry.
• SKILL.md Compatibility — Fully compatible with the Agent Skills standard used by Claude Code, GitHub Copilot, and OpenAI Codex.
• Demo Agents — Includes ready-to-run examples for code review, PDF processing, SEO auditing, data analysis, email drafting, and web scraping.