Code-Mode is a plug-and-play library that lets AI agents call tools by executing TypeScript (or via a Python wrapper) instead of making many individual function calls. Its core philosophy is that language models are very good at writing code, so rather than exposing hundreds of separate tool endpoints, you give the model a single “code execution” tool that has access to your full toolkit through code. This approach can dramatically reduce the number of tool-call iterations needed in complex workflows, turning multi-step call chains into a single code execution with internal branching and loops. The repository contains both TypeScript and Python libraries, plus a code-mode-mcp component for integrating with MCP and UTCP ecosystems. Benchmarks in the README highlight improvements in latency and token cost for scenarios involving multiple tools, showing that code execution often outperforms traditional JSON-based function calling.
Features
- Single “code execution” tool interface that replaces dozens of individual tool calls
- TypeScript and Python libraries for integrating Code-Mode into different agent stacks
- Code-mode-mcp bridge for connecting to MCP and UTCP tools through one entry point
- Benchmarked improvements in latency and token usage for multi-tool workflows
- Flexible architecture that lets LLMs orchestrate conditionals, loops and branching inside code
- MPL 2.0 licensed project suitable for commercial and open deployments