Most configuration decisions can be revisited; this one cannot, so a server created with public OAuth for convenience stays that way. The documents tool being disabled by default is the more interesting default: retrieval returns passages, and handing over whole documents is an extra decision rather than an assumption. Customisable tool names and source-group scoping mean one Kapa project can front several differently-scoped servers.
Kapa's per-customer servers, each on its own subdomain. You create the integration, choose the subdomain and the authentication method, and customise what the tools are called and which source groups they cover.
- Search_<product>_knowledge_sources — agentic retrieval returning passages with source URLs
- Get_<product>_knowledge_documents — full documents by URL; off by default and enabled deliberately
- give_feedback
- Tool names, descriptions and the source groups they cover are all configurable
A Kapa project. The authentication method is set when the server is created and cannot be changed afterwards — choose it deliberately.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen