Asking an AI to "add a conclusion" used to mean regenerating the entire article. Every paragraph came back through the model, you waited proportionally longer, and small wording changes crept in where you never asked for them.
Tilnote's MCP connector solves this with two tools that work as a pair.
Read the structure first
get_note_blocks splits a note into addressable blocks and hands back an outline:
B1 heading line 1 Why indexes stop helping
B2 paragraph lines 3-7 Every index you add...
B3 code lines 9-14 db.pages.createIndex(...)Along with the blocks you get a revision value. Hold onto it.
Then change only what moves
edit_note takes a list of operations, each pointing at a block ID:
replace— swap one block's contentinsert_after/insert_before— add around a blockdelete— remove a blockappend/prepend— add at either end
Only the blocks you name are touched. Everything else is byte-for-byte unchanged, which is the actual point: your prose stays exactly as you wrote it.
Why the revision is required
Block IDs are positional. B3 means "the third block right now" — so if the document shifts between reading and writing, an edit aimed at B3 lands on the wrong paragraph.
That is why revision is mandatory rather than optional. The write only lands if the note still matches what you read. If you had the web editor open in another tab and saved something in between, the edit is rejected with the current revision instead of silently overwriting your work.
When an edit succeeds you get a fresh revision back, plus a reminder that the block IDs have shifted and need re-reading before the next round.
When to use which
Use edit_note for surgical changes to a document you want preserved. Use update_note when you genuinely intend to replace the body wholesale — it is faster to reason about, but it is a full overwrite, so treat it accordingly.
For simply tacking something onto the end, append_to_note is simpler than either.
이 노트는 보도·비평·교육·연구 목적으로 작성되었습니다. 저작권 문의가 있으시면 에서 알려주세요.
