No reflexivity about it's datasets

[tdelabro]

❯ Quels sont les dataset dont tu disposes?

⏺ listMcpResources(List MCP resources from server "datagouv")
  ⎿  (No resources found)

⏺ Je n'ai pas de datasets pré-chargés. Le MCP data.gouv.fr me donne accès à des outils de recherche et d'accès, pas à une liste fixe de datasets.

  En pratique, je peux interroger l'ensemble du catalogue de data.gouv.fr à la demande via les outils search_datasets et search_dataservices. La limite est que si les données n'ont pas été publiées en open data sur cette plateforme, je ne peux pas y accéder.

  Dis-moi ce que tu cherches, je fais la recherche !

I would like for claude to be able to tell me what dataset it has access too.
I keep asking stuff for it to tell me it does not have access to this data, eg.

Les données de l'Assemblée nationale ne sont pas publiées sur data.gouv.fr. Cependant, l'Assemblée nationale dispose de sa propre open data sur data.assemblee-nationale.fr, mais cela dépasse le périmètre du MCP dont je dispose.

  Pour retrouver les votes d'Antoine Léaument directement :

  - Fiche député : assemblee-nationale.fr → recherche par nom
  - Scrutins publics : chaque vote en séance plénière est publié avec le détail par député

But it can't tell me which data it has access to

[bolinocroustibat]

data.gouv.fr has 70,000+ datasets available, they can't be loaded in the LLM context - but we could think about having a tool to request the catalog cc @estellebertrand

[estellebertrand]

I believe that might actually be a good case to add a resource to the server about it. We could describe a bit more what type of data or data services can be found on Data Gouv, and some other important information about the catalog itself. If the agent is confused, it can then call the resource to retrieve it and add it to the context. What do you think about this @bolinocroustibat ?
Unless you have an idea of good catalog tool that could be used, that might be a good start !

[bolinocroustibat]

Related: #71

[bolinocroustibat]

I believe that might actually be a good case to add a resource to the server about it. We could describe a bit more what type of data or data services can be found on Data Gouv, and some other important information about the catalog itself. If the agent is confused, it can then call the resource to retrieve it and add it to the context. What do you think about this @bolinocroustibat ? Unless you have an idea of good catalog tool that could be used, that might be a good start !

We could use https://www.data.gouv.fr/datasets/catalogue-des-donnees-de-data-gouv-fr in a specific new tool (getting the latest catalog resource, then browing TabularAPI for it for example), and/or in a docstring? Just an idea

[estellebertrand]

We don't have tools for each of the item listed on this page no? I'm afraid it might confuse the agent if it reads "La liste des tags créés sur data.gouv.fr" but then can't get this tags list (i don't know enough the current possibilities with the api x tools integrated) ?

[bolinocroustibat]

We don't have tools for each of the item listed on this page no? I'm afraid it might confuse the agent if it reads "La liste des tags créés sur data.gouv.fr" but then can't get this tags list (i don't know enough the current possibilities with the api x tools integrated) ?

Main API has an endpoint to list the ressources (files) of a dataset (set of files) (e.g. here and here), so we could have a tool (or a process in an existing tool) to get the catalog of datasets/dataservices/reuses easily and reliably using this endpoint and a simple regex to get the right catalog file, for example

[bolinocroustibat]

This ticket #92 has been created as a first step, after discussion with @estellebertrand