Skip to main content
Search a vault and return the top-k full notes in one call, instead of a search followed by read_notes. Ranking is vault_graph_search’s, limited to notes you can read. mode=note (default) returns each note’s frontmatter and body; mode=section returns the heading section each hit matched. Every item carries chunk_id, the chunk it was ranked by (pass it to record_retrieval_feedback). k is at most 20. The result is held under the server’s byte budget, shared equally across the notes: a note over its share is cut and marked truncated: true with size_bytes (its full size); fetch it whole with read_note. Anything that still does not fit comes back with next_cursor: repeat the same call plus cursor until it is null. An item that cannot be returned is a per-item error with its rank (a missing note and an unreadable one look the same). A cursor is bound to the caller, the tool and these exact arguments, and expires. response_format=concise returns {path, rank, score, body, content_hash, chunk_id} per note, without frontmatter in note mode; size_bytes and truncated appear only on a truncated item, and section_resolved only when false.

Annotations

Idempotent is the advisory MCP idempotentHint. Destructive is also yes for a tool that asks for confirmation only on some calls.

Confirmation (human in the loop)

No human confirmation is required to call this tool.

Input

Output

Other tools in the search domain: query_datacore, search, search_dql, search_jsonlogic, search_omnisearch, search_regex, search_semantic, search_text, search_vault. Back to the tool catalog.