AgentEval

About

Addresses Julia's notorious 'Time to First X' (TTFX) problem, which significantly slows down AI agent workflows by incurring 1-2 seconds of startup time for every command. This tool provides a persistent Julia session using the Model Context Protocol (MCP) via STDIO, allowing AI agents like Claude Code to keep a Julia process alive. This means the overhead of startup, package loading, and JIT compilation is paid only once, leading to dramatically faster and more efficient interactive Julia code execution for AI assistants.

Key Features

• STDIO transport for enhanced security and no network port requirement
• Automatic server startup by AI agents when needed
• Persistent Julia session state, maintaining variables, functions, and loaded packages across calls
• Eliminates Julia's 'Time to First X' (TTFX) overhead for subsequent code evaluations
• Eligible for inclusion in the Julia General registry for wider distribution
• 1 GitHub stars

Use Cases

• Developing AI agents that require interactive and stateful Julia environments
• Automating Julia code execution within AI-driven development and analysis workflows
• Enabling AI coding assistants to execute Julia code efficiently and persistently