Development with a Dot

Introduction Some of you may remember - and hopefully have even tried! - my Isolator project . Isolator is a framework for running isolated code for .NET. What this means is that possibly unsafe code can be run in a secure, isolated, way. It offers some quite different isolation strategies: Process isolation : the code to execute runs in another process, possibly under a different identity (on Windows only) Assembly isolation : uses an assembly load context to isolate the code execution, which is then unloaded Distributed isolation : the code to execute is sent to possibly another server for execution Scanned : the code to execute is first checked for problematic code using my ReferencesScanner project Now, I added support for Docker isolation! What this means is, if you have Docker running on your machine, you can spawn an image that will just be used for running your code, and after that will be gone. This provides another very restricted level of isolation. For Docker A...

Introduction This post picks up where this one left. It will be my second post on using LLMs, and AI, in general. This time I'm going to cover integrating MCP tools with the LLM's response. MCP stands for Model Context Protocol , and it is an open standard . In a nutshell, it is a protocol designed to help LLMs communicating with the real world, for example, accessing a database, getting real-time weather information for a specific location, creating a ticket in some system, sending out an email, etc. We need an MCP host and some tools registered with it. There are many ways by which LLMs can communicate with the MCP host, always using JSON-RPC 2.0 for messaging: Standard input/output, if running on the same machine HTTP calls Streams, mainly for the same machine Custom-defined Now, I won't go through all of them now, I'll just pick HTTP transport, as it's probably the most usual one. Also, I will be using the  OpenAI  API. Essentially, we registe...