PxWEB and MCP

[Uldis Ainārs]

Hi all,

At CSB Latvia we’re considering using the MCP for other projects and are keen to know if there are any current or future plans regarding PxWEB and MCP development.

Any insights would be greatly appreciated.

Best regards,
Uldis Ainārs
CSB Latvia

[Lars Pedersen]

Hi Uldis,

 

This code uses the Danish Statbank api, still slightly different from Pxweb for mcp

 

And it is in Danish

 

https://github.com/brokk-sindre/mcp-dst

 

So is this short demo video

 

https://www.linkedin.com/posts/mikkelkrogsholm_ai-claude-automation-ugcPost-7310962122857889792-uipu?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAABpNCkBUCwQV-Yn0oAtCDu7FY_csjIV5M0

 

 

But you might find It interesting

 

J Lars

--
You received this message because you are subscribed to the Google Groups "pcaxis" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pcaxis+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/pcaxis/92b95812-5305-4356-8654-e3099d8b2cb8n%40googlegroups.com.

[Petros Likidis]

Hi,

We talk about it but we do not have any concrete plans to do anything that I know.

Kind regards

Petros

[Uldis Ainārs]

Hi
You can now play around with the MCP endpoint at:
https://arturslunevs-csplv-mcp-test.hf.space/gradio_api/mcp/sse
using data from stat.gov.lv.

A supporting Python script is available here: https://huggingface.co/spaces/arturslunevs/CSPLV_mcp_test/blob/main/app.py
It demonstrates how to connect, call functions, and handle responses from the API.

One challenge when working with LLMs is that PX-Web often provides separate JSON arrays for codes and names. This makes it harder for the model to align which code belongs to which description. The new get_query_values function solves this by returning a dictionary {code: name}, which is much clearer and easier to use

Here AI story about foreign trade - https://claude.ai/public/artifacts/7d4e6a8d-59c6-4fe6-b66c-a33bb2aea74e

Uldis.