Module code_executor 

Code execution environment for agents using MCP tools programmatically.

This module allows agents to write and execute code snippets that interact with MCP tools as library functions, rather than making individual tool calls. This improves efficiency through:

• Control flow: loops, conditionals, error handling without repeated model calls
• Data filtering: process results before returning to model
• Latency: code runs locally with direct process execution
• Context: intermediate results stay local unless explicitly logged

Example

let executor = CodeExecutor::new(
    Language::Python3,
    Arc::new(mcp_client),
    PathBuf::from("/workspace"),
);

// Agent writes Python code
let code = r#"
files = list_files(path="/workspace", recursive=True)
filtered = [f for f in files if "test" in f]
result = {"count": len(filtered), "files": filtered[:10]}
"#;

let result = executor.execute(code).await?;

Structs

BuiltinToolInfo

A built-in vtcode tool exposed to agent code snippets, mirroring the MCP tool surface so the model can call built-in tools as library functions and filter/aggregate data in-process.

CodeExecutor

Code executor for running agent code.

ExecutionConfig

Configuration for code execution.

ExecutionResult

Result of code execution.

Enums

Language

Supported languages for code execution.

Traits

BuiltinToolExecutor

Executor that runs built-in vtcode tools from code snippets.