Description

It’s no secret that AI runs the world now, either because of hype or because of its true potential. People no longer search in Google, but chat with LLMs. At the moment of writing, the main restriction of LLMs is their lack of connectivity with the outer world and external systems. That’s where the Model Control Protocol (MCP), and MCP servers fill the gap. MCP is born as an interface to connect AI models with the outer world.

Application Collection is the curated, trusted, and up-to-date collection of developer and infrastructure applications built, packaged, tested, and distributed by SUSE. Although this product has already a couple of years now, some people still struggle to integrate it into their technology stack: a developer might not know how to change a Dockerfile from public images to the Application Collection ones, infrastructure engineers wonder whether if they can replace 1:1 a public Helm chart with one from AppCo, or a customer success engineer needs to know how to start using a given application.

With an MCP server, most of this questions are answered instantly by a large language model:

• The developer will command the model to migrate their current Dockerfile to use images from Application Collection. The model will then search the collection, look for the latest releases, and apply the necessary changes whenever related images are found.
• The infrastructure engineer will ask the model to generate a migration guide for a Helm Chart already installed in their cluster. The model will look for the internal chart documentation and will help the user during the migration: generating backups, configuring authentication or preparing the new release.
• The customer success engineer will ask the LLM to generate a Hello World demo for a new app, and the model will just do it ready for the next meeting.

Goals

• Learn about MCP servers: state of art, frameworks, good design practices, etc.
• Learn about AI agents: model types, sizes, frameworks, existing solutions
• Develop a basic MCP server with enough tools to address the use cases mentioned above
• Evaluate whether if this can be a new product line

Resources

• Application Collection REST API specification
• Application Collection documentation site
• Model Context Protocol site
• VSCode Copilot MCP servers support
• Claude Desktop MCP servers support
• LangChain agents development framework
• Anthropic: Writing effective tools for AI agents
• OpenAI: A practical guide to building agents