Patterns for "Code Execution" & Agentic Scripting with the C# SDK

[tadodev]

Pre-submission Checklist

• I have checked that this question would not be more appropriate as an issue in a specific repository
• I have searched existing discussions and documentation for answers

Question Category

• Protocol Specification
• SDK Usage
• Server Implementation
• General Implementation
• Documentation
• Other

Your Question

I've been reading recent engineering posts from Anthropic and Cloudflare regarding the shift from standard tool-calling loops to "Code Mode" (Code Execution).

The core idea is moving away from the chat-based loop (LLM -> Tool Call -> Response -> LLM) and towards letting agents write and execute scripts that use MCP tools as importable libraries. This approach significantly reduces latency and token usage for data-heavy tasks.

My Question: Do we currently have any examples or best practices for implementing this pattern using the C# MCP SDK?

Thank you

[PederHP]

You need two things to do code mode mcp:

1. code execution tool(s), which could be mcp based

2. a code execution runtime which has either virtualized mcp client capabilities or is an actual full code execution runtime.

I do not recommend you build this yourself, unless that is your project focus or as a learning experience. It's not trivial and it certainly isn't easily retrofitted onto an SDK.

[CharlieDigital]

I have a project that enables this using Jint: https://github.com/CharlieDigital/runjs

Effectively it allows the LLM to generate and execute JavaScript over MCP in a safe sandbox.

With Jint, you can also expose your runtime .NET domain to it.