Giving your AI agent access to up to date :github_...
# ai-cafe
m
Giving your AI agent access to up to date github on fire GitHub knowledge and endpoints is nice, but the complexity of GitHub actions is nothing against the plethora of aws AWS services, configurations and policies. Therefore, I am really happy about the AWS MCP server - maybe AI will understand what clearly escapes my mind. ๐Ÿ˜„ On the other hand, it seems odd, because poorly configured and oversized AWS services probably constitute a significant share of their income. ๐Ÿ˜ฌ
๐Ÿ˜ 4
f
Excited to try this out! Let us know about your experience with it @Matthias Zepper!
m
I currently have no use for it. But I included MCP into my AI introduction a while ago and possibly (long-term goal) want to make some of our LIMS data available via a private MCP endpoint for my colleagues to use. But currently AI adoption is still an uphill battle at NGI (except for Cursor and Copilot).
๐Ÿ‘ 2
m
did you follow HuggingFace's agent course ? @Matthias Zepper
m
I skimmed over it, but I didnโ€™t take it, no. I am aware of the basic concepts of agents (I follow the developments out of interest and two months ago gave a lecture at SciLifeLab about LLMs) but my job is not compatible with building an AI application myself. I donโ€™t have the time to do it outside the job and no use for it inside the job. Maybe some day we will agree on adding vector search or an MCP endpoint on our databases. But even in that case, the consumers of the endpoint will just be third-party applications like local Claude, Goose, Starlight Search or whatever. Basically every day I have a new agent framework or AI tool capable of using MCP endpoints in the GitHub feed, so there are enough other people building it. ๐Ÿ˜„
๐Ÿ‘ 1
f
I did some very crude experimentation with MCP for nf-core tools yesterday. I started by directly calling the CLI and then tried using the python package directly but ran into some issues with arguments passed to
ModuleCreate
. The interactivity of many commands make them a bit difficult to implement in MCPs (at least for a novice like me). But was a good learning experience to understand how to get started with MCP development! โš ๏ธ Personal experimentation, don't use for production: https://github.com/FloWuenne/mcp-nf-core
nf core scroll 1
๐Ÿค– 1