agent-notion

Two backends, one surface — the official REST API and the v3 desktop session; --backend auto prefers a stored v3 session and falls back to the official credential

Full CRUD — search, read, create, update, trash, restore and archive pages, databases and blocks

Markdown in and out — page content renders as markdown with GitHub-flavored pipe tables, and content is appended or replaced from markdown

Notion AI — list models and threads, send messages against a page, stream responses, mark threads read

Export — a page (optionally recursive) or the whole workspace, as markdown or HTML, with task polling

Inline comments anchored to text — target a phrase by content and occurrence, not by offset

Backlinks, version history and the activity log — the questions the REST API cannot answer

Warns where Notion will mislead you — database query returns the view's own order, often manual, so --limit can silently omit a freshly-added row; the docs tell you to sort on a timestamp and watch the @pagination trailer

Tokens stay off argv — auth import reads from stdin, validates against the API before storing, and derives the alias from the workspace name

Explicit guidance for agent-driven setup — an agent should tell the user to run the import themselves rather than accept a pasted token on --token

Destructive commands refuse without --yes — trash, archive, block delete, block replace, logout

Multi-workspace — named aliases, switch and set-default, keychain-first with a warning field when it falls back to plaintext

Zero runtime dependencies — a single static Go binary