took me a bit and then it clicked. you connect MCP tools which are stdio jsonrpc servers. they provide a tools/list endpoint which goose loads into the ai prompts so that they know what to do.

there's not much more to it, goose also provides ways to connect to different ai backends.

enabling the developer mcp is most useful, since it gives ais access to your shell

Reply to this note

Please Login to reply.

Discussion

Yeah, i’m muddling through it… i wasn’t familiar with MCP and i think there’s a few things you need to do in setting up goose to make it actually work well.. Documentation and improve tooling will help. My suspicion is that the goose developers are so embedded in their world that they can’t see what others need to do to get started.

here’s a cli I was using before Goose showed up. same ux but a little more obvious how it plugs together.

https://github.com/mark3labs/mcphost

can it also control the shell inside a container? would be cool to tell the ai to quickly spin up a container with some software and then let it work with that