---
type: Changelog Entry
title: Permissions in plain words instead of tool checkboxes
resource: https://www.librario.de/en/changelog/2026-09-berechtigungen
tags:
- changelog
---

# Permissions in plain words instead of tool checkboxes

Connecting an AI assistant to Librario used to put a list of tool names in front of you, so deciding what to allow meant knowing what `download_file` or `get_publication` actually do. The consent screen now shows one permission per line in plain words: "Search and view publications" and "Download and read attached files". The second one combines what used to be two separate checkboxes, one for downloading and one for reading file content. What each checkbox does is spelled out in the [AI assistants documentation](/en/docs/integration/mcp).

**Already-connected AI assistants:** if you had allowed a connection only one of the two, downloading or reading file content, it does not get the combined permission automatically, since that would have allowed it more than you had consented to. It therefore loses the file capability it did have until you reconnect it under Account → Integrations and grant the permission again. Connections you had allowed both, or neither, keep exactly the access you had given them.

The REST API knows the same distinction now as `assets:read` and `assets:write`, instead of a shared `library:write` for files. REST API tokens issued before this change received `assets:read` or `assets:write` automatically wherever they carried `library:read` or `library:write`; AI-assistant connections follow the rule above instead. New integrations need `assets:write` explicitly to upload, change, or delete files; `library:write` alone is no longer enough for that. Which scope opens which endpoints is listed in the [API documentation](/en/docs/integration/api).
