Skillscript

Overview

Skillscript is a declarative, sandboxed language for tool orchestration, positioning itself as an MCP-native platform where "the skillscript is the program; the model is the runtime." Released pre-1.0 and self-hosted, it provides a structured way to define tool workflows that AI agents can execute safely within a sandbox environment. The project emerged as part of the growing ecosystem of MCP-native development tools designed to give coding agents structured, declarative control over tool usage.

Features

• Declarative orchestration: Define tool sequences and dependencies in a purpose-built language
• Sandboxed execution: All tool calls run in an isolated, secure environment
• MCP-native: Built on the Model Context Protocol for seamless agent integration
• Self-hosted: Run your own Skillscript runtime without cloud dependencies
• Model-agnostic runtime: The AI model acts as the runtime, interpreting and executing skillscripts

Installation

# Skillscript is currently pre-1.0
# Install via npm (check official docs for latest)
npm install -g @skillscript/cli

Configuration

{
  "mcpServers": {
    "skillscript": {
      "command": "skillscript",
      "args": ["serve"]
    }
  }
}

Usage Examples

Skillscript allows you to define orchestrations declaratively:

// Example: Research workflow
workflow research {
  step1: search(query="latest AI agents")
  step2: fetch(urls=step1.results[:3])
  step3: summarize(content=step2.data)
}

Pros

• ✅ Clean declarative syntax for tool orchestration
• ✅ Sandboxed execution for safety
• ✅ MCP-native design
• ✅ Self-hosted option for privacy

Cons

• ❌ Pre-1.0, API still evolving
• ❌ Limited documentation compared to mature projects
• ❌ Smaller community

When to Use

• You need structured, declarative tool orchestration for AI agents
• You want MCP-native development with sandboxed execution
• You're building agent tool workflows and want a purpose-built language

Resources

• GitHub
• Documentation