Tools to support "Get this setup on Buildkite"

[toolmantim]

I went to use this with https://ampcode.com/ in the hope of saying "Get this repo setup on Buildkite" and have it do all the work for me (including verifying/iterating until things work). Bonus points for asking it to iterate on the Docker/BK config for maximum performance (i.e. getting the right Docker caching work). At a glance though the tools are quite generic? Are there example use cases they're designed to support?

In the meantime, I've just opened buildkite/cli#478 so it can be done using the CLI instead

[mcncl]

@toolmantim 👋 at the moment the MCP server only has read capabilities, so in its current state wouldn't be able to get your from 0 to 1 in terms of a Buildkite set up. Really keen on getting it to that point, but we wanted to get a sense that folks are using it and that it adds value to workflows before committing too much time to it.

Under the hood it's using go-buildkite, so anything which can be done in there can be done via the MCP, assuming access is granted via prompt etc.

In terms of get me set up on Buildkite, we'd only need to add a write_pipelines permission to the API key, and a buildkite.pipeline.Create to the tools list.

[toolmantim]

Thanks! I'll see if I can get a dev environment setup for this and have a play.

I was hoping for some examples at least for what these MCP tools are best used for today, so I could experience anything great you've got working already.

[mcncl]

@toolmantim sure thing! Can definitely provide some examples of how we've seen it used, and how we use it. We're in the process of improving log processing, given how many tokens they can take up.

[mcncl]

@toolmantim at the moment, we have this set up as an assistant for debugging/getting information on the current state, including things like build performance.

We've used it as a can you give me a sense as to what caused X build to fail?, tell me how my builds perform on average, are there any outliers? why? and so on. We're making steps to allow a full 0-1 set up with additions like #34, so working with something like Amp you'll be able to set up your project to run in CI. We can point it to the schema to lint the pipeline file as it goes, which should help avoid errors about incorrect types.