← ClaudeAtlas

voicenotes-pulllisted

Pull voicenotes for a user-specified time window OR tag, AND surface notes that `list_notes` alone silently drops. Use whenever the user asks "what voicenotes did I make from X to Y", "pull my notes about Z", "get my notes tagged X", or otherwise expects completeness over a time range or a tag.
vibhasjain/agent-keyboard · ★ 5 · AI & Automation · score 70
Install: claude install-skill vibhasjain/agent-keyboard
# voicenotes-pull The voicenotes MCP exposes four tools (`list_notes`, `search_notes`, `get_transcript`, `create_text_note`) but **`list_notes` is buggy: it silently drops notes that exist on the server and fall inside its `date_range`**. In one observed case it returned 5 of 9 notes in the requested window even with `limit: 50` and a tight time range. If you only call `list_notes` and report the result, you will miss notes the user knows exist. They will push back. Don't get caught. ## The workflow When the user asks for notes over a time window (or implies completeness with words like "all", "everything", "the bunch", or names a date range) — OR asks for notes **by tag** ("get my notes tagged mobile", "the meeting notes"): ### Step 1 — list pass Call `mcp__voicenotes__list_notes` once with: - `date_range: [<startISO>, <endISO>]` covering the requested window. Dates are UTC. If the user said "yesterday evening" and you're in EDT, convert: yesterday evening EDT = 23:00–04:00 UTC across that date boundary. - `tags: ["<tag>"]` when the request is tag-scoped (lowercase the tag; the API matches case-insensitively). - `limit: 50` (default is 10, almost never enough). Collect every UUID into a set. Don't show the user yet. ### Step 2 — search sweeps (the part that catches the dropped notes) In **one message, parallel tool calls**, run **3–5 `mcp__voicenotes__search_notes` calls** with keyword variations covering what the notes are likely about. Two sources for keywords: 1