MCP servers are a really useful way to extend the capabilities of a large language model running locally. They can provide access to a wide range of APIs and data. I've made it easy to write your own in Xojo with my MCPKit module but I also have several that I've developed myself which are listed below.

mcpfetch

An MCP server that allows the LLM to provide a URL and request either the raw HTML or a Markdown summary of the page contents.

https://github.com/gkjpettet/mcpfetch

mcpkagi

This allows your LLM to search the web using the excellent Kagi search engine. It requires a Kagi API key to function. This should be provided to the LLM client with the command line --apikey argument.

https://github.com/gkjpettet/mcpkagi

mcpwikipedia

Gives your LLM the ability to search Wikipedia using the official Wikipedia API. You should pass to the server (via the command line argument --useragent) a unique User Agent string for this tool.

https://github.com/gkjpettet/mcpwikipedia