[Discussion] MCP for Discoverability

[pdelboca]

I'm wondering if the main mcp-ckan repository should held generic tools for the CKAN API.

I know it is not the main focus of the project but, as a product, is a relatively quick win. AI is good for discoverability issues and it doesn't pose a treat since we do not have the veracity/exactness problem. I'm seeing a huge benefit since it will give a boost of UX to the tool, at the same time I don't want it to divest it from our focus.

I'm starting this issue to see how much of a quick win it is.

Some Tech Thoughts

Ideally we will have something similar to the OnData server or any other CKAN MCP server, a wrapper to the CKAN API that the LLM can call to discover datasets.

The part specifc to this server is the registration of a new CKAN Data Portal to query. Since we are registering tools and data-portals through Python Packages, we can as well register new portals using the package as well.

There is an orchestation part of it. If we have several plugins installed with different portals, we need a package_search to each registered data portal.

[avdata99]

Proposal: This probably can be a new mcp-tool repo like other we already created.