Code Mode
Let agents call MCP & UTCP tools via code execution
Code Mode – Execute complex tool workflows via TypeScript sandbox
Summary: Code Mode enables large language models (LLMs) to call MCP and UTCP tools by executing TypeScript code within a sandbox, replacing traditional tool calls. This approach consolidates multi-step workflows into single batched executions, improving efficiency and reducing token usage significantly.
What it does
Code Mode provides a TypeScript execution sandbox that grants LLMs access to an entire toolkit, allowing them to write code to invoke tools instead of relying on multiple tool schemas. This streamlines complex workflows into one execution.
Who it's for
It is designed for developers and teams integrating LLMs with MCP and UTCP tools who need efficient, scalable multi-tool workflows.
Why it matters
It addresses LLMs’ difficulty with tool calling by reducing execution time, token consumption, and round-trips, enhancing performance in complex agent workflows.