Amazon Quick Desktop: LMA Workflows

Amazon Quick Desktop: LMA Workflows

Once LMA is connected to Amazon Quick Desktop as an MCP server, the seven LMA tools become first-class capabilities that Quick can compose with any other tool you’ve connected — your calendar, Slack, task trackers, the knowledge graph, web search, and other MCP servers. That composability is what makes the integration powerful: it isn’t a fixed feature set, it’s a building block.

This guide shows three layers of usage:

1. Conversational workflows — ad-hoc prompts you can type into Quick Desktop chat today, with no setup beyond the MCP connection. These are the easiest way to get value.
2. Scheduled agent recipes — agent configurations that run autonomously on a schedule or trigger (pre-meeting briefings, action-item extraction, KG enrichment).
3. The skills pack — a packaged bundle of the most popular workflows, installable in one step.

First time setting up? Connect LMA to Quick Desktop first — see Amazon Quick MCP Setup → Path B. Or, for the fastest install, use the amazon-quick-desktop-skills-pack/ bundle which ships agents and skills as a one-step install.

Prerequisites

• LMA connected to Quick Desktop via MCP API key (see Amazon Quick MCP Setup)
• Quick Desktop with calendar (Outlook or Google), and optionally Slack and a task tool (Asana, Jira) connected
• Verified that LMA tools work in chat — try Search my LMA meetings for [a recent topic] and confirm results

Available LMA tools

LMA exposes seven tools, of which five are read-only (annotated readOnlyHint: true so Quick Desktop classifies them correctly for scheduled agents — no allow_unscoped_write_tools required for read-only flows):

• Read-only: search_lma_meetings, get_meeting_transcript, get_meeting_summary, list_meetings, get_virtual_participant_status
• Write: schedule_meeting, start_meeting_now

All responses include meetingUrl and (where applicable) virtualParticipantUrl deep-links so agents can post clickable references back into the LMA UI.

---

Conversational workflows

Once LMA is connected, Quick Desktop will automatically pick the right tools to satisfy your request — including chaining LMA with your calendar, Slack, task tracker, knowledge graph, web search, and any other MCP servers you’ve installed. You don’t need to write an agent or a skill; just ask.

The examples below all work as plain chat prompts. Each one shows which tools Quick will typically reach for so you can see the composition.

Join my next meeting

“Send the LMA virtual participant to my next scheduled meeting.”

Quick reads the next event from your calendar (Outlook/Google), pulls the meeting URL from the invite body, and calls live_meeting_assistant_lma__start_meeting_now with the right name and URL — then polls get_virtual_participant_status until the VP is live and reports back with the VNC viewer link.

Variants that work the same way:

“Join my 2pm with the design team — name the participant ‘Design Sync Notes’.”

“Schedule the LMA bot for every recurring 1:1 on my calendar this week.” (uses schedule_meeting)

“If my next call is on Zoom and I have stored Zoom credentials, sign in — otherwise join as a guest.”

Pull action items from the last meeting

“What did I commit to in my last meeting? Post the action items assigned to me into #my-todos in Slack.”

Quick calls list_meetings to find the most recent one, get_meeting_summary with includeActionItems=true, filters for items owned by you, and posts to Slack with the LMA meetingUrl deep-link so your team can click through to the source transcript.

Replace Slack with whatever you have connected:

“…create Asana tasks in the Q3 Launch project for each action item assigned to me, with the meeting link in the description.”

“…add them as Jira tickets under epic LMA-42 and tag the owner.”

”…DM each owner in Slack with their items.”

Answer questions from past transcripts

LMA’s transcripts are searchable knowledge in their own right.

“Did we ever decide on the database vendor for Project Atlas? Search my LMA meetings and quote the relevant transcript line with the meeting link.”

Quick calls search_lma_meetings with topical queries, then get_meeting_transcript on the top hits to find the literal quote, and returns a citation with meetingUrl.

More examples:

“Find every meeting where Priya mentioned the budget cap and summarize her position over time.”

“What was the timeline we agreed with the vendor in last Thursday’s review? Quote the transcript.”

“I’m about to email the customer — pull every commitment we’ve made to them across our last five calls.”

Post a recap to Slack / email / Confluence

“Summarize this morning’s all-hands and post it to #general with the top three decisions and any unresolved questions.”

Quick locates the meeting via list_meetings, calls get_meeting_summary, shapes the output for Slack, and posts. Works the same with email, Confluence, Notion, or any other connected destination.

“Email the recap of today’s customer call to person@example.com with action items as a checklist and the LMA recording link at the top.”

“Append the decisions from this week’s design reviews to the Design Decisions page in Confluence.”

Pre-meeting prep on demand

You don’t need a scheduled agent to get a brief — just ask before the meeting:

“I have a call with Acme Corp in 30 minutes. Pull every prior LMA meeting with anyone from acme.com, summarize what we last discussed, and list any open action items.”

Quick uses your calendar to identify attendees, runs search_lma_meetings, calls get_meeting_summary on the top results, and produces a brief in the chat — with one-click meetingUrl links to each source meeting.

Cross-tool synthesis

The real win is when LMA is just one source among many:

“Cross-reference the action items from yesterday’s standup with the open Jira tickets in sprint 42. Anything we committed to but haven’t ticketed yet?”

“From the last four customer calls, extract every feature request and match it against open issues in our GitHub repo. Report any gaps.”

“Look at the action items from this morning’s exec sync and check Slack #leadership for any follow-up discussion in the last 24h. Summarize the current state of each item.”

These prompts blend search_lma_meetings / get_meeting_summary with Jira, GitHub, and Slack tools — Quick handles the orchestration.

Tip: use the deep-links

Every LMA tool response includes a meetingUrl (and virtualParticipantUrl for VP launches). When you ask Quick to post, email, or DM, tell it to include the link — that turns every output into a clickable reference back into the LMA UI for the full transcript, recording, or live VP viewer.

---

Scheduled agent recipes

The patterns above are great for ad-hoc use. When you want LMA workflows to run automatically on a trigger or schedule — without you typing a prompt — define them as Quick Desktop agents. The recipes below are ready-to-paste configurations.

Recipe 1: Pre-Meeting Briefing Agent

Runs every 5 minutes; fires 15 minutes before any calendar event. Searches LMA for prior meetings with the same attendees, pulls summaries and action items, cross-references the knowledge graph, and posts a concise brief to your activity feed.

Configuration

Agent ID: lma-pre-meeting-brief
Model: smart
Schedule: interval (every 5 minutes)

Tool policy:

[
  {"group": "outlook_builtin"},
  {"group": "user_mcp__live_meeting_assistant_lma"},
  {"group": "kg_tools"},
  {"effect": "allow", "tool": "live_meeting_assistant_lma__search_lma_meetings"},
  {"effect": "allow", "tool": "live_meeting_assistant_lma__get_meeting_summary"},
  {"effect": "allow", "tool": "live_meeting_assistant_lma__list_meetings"},
  {"effect": "allow", "tool": "live_meeting_assistant_lma__get_meeting_transcript"}
]

Trigger:

Template: on_upcoming_event
Trigger ID: upcoming-meeting
Parameters: {"minutes_ahead": 15, "include_all_day": false}

Agent prompt:

You are a pre-meeting briefing agent. Your job is to prepare concise, actionable
briefings before upcoming meetings by combining calendar context with LMA
meeting history.

## When triggered

You receive context about an upcoming calendar event (subject, attendees,
time). Your task:

1. Identify the meeting participants from the trigger context.

2. Search LMA for prior meetings with those participants:
   - Call `live_meeting_assistant_lma__search_lma_meetings` with queries
     like the participant names or the meeting subject/topic.
   - If results are found, call `live_meeting_assistant_lma__get_meeting_summary`
     on the most recent 1–2 relevant meetings to get summaries and action
     items.

3. Check the knowledge graph for relationship context about the attendees
   (recent interactions, commitments, project associations) using `kg_search`.

4. Synthesize a brief that includes:
   - Meeting: subject, time, attendees
   - Last interaction: when you last met with these people and key topics
   - Open action items: any outstanding commitments from prior meetings
   - Context: relevant relationship/project notes from the knowledge graph
   - Preparation suggestions: what the user might want to review or prepare

5. Post the brief to the activity feed using `update_feed` with:
   - importance: "important" (so it shows a toast notification)
   - A clear, scannable format with bullet points
   - Include the LMA meetingUrl from each referenced meeting so the user can
     jump to the full transcript with one click.

## Guidelines
- Keep briefs concise — aim for 3–5 bullet points, not a wall of text.
- If no prior LMA meetings are found with the attendees, say so and still
  provide any KG context.
- Skip trivial/recurring meetings with no meaningful history.
- Never fabricate information — only report what you actually find in LMA
  and the knowledge graph.
- Do NOT call schedule_meeting or start_meeting_now — you are read-only.

Test, then enable

After creating the agent, do a dry run before turning on the schedule. Quick Desktop accepts natural-language requests for agent management — ask it to run the agent once, review the output, then ask it to enable the schedule. For example:

“Run the lma-pre-meeting-brief agent now so I can see the output.”

(once you’re satisfied)

“Enable the lma-pre-meeting-brief agent on its schedule.”

You can also drive these actions from the Quick Desktop Agents panel (Settings → Agents) if you prefer a UI gesture to a chat prompt. Once enabled, the agent runs every 5 minutes, checking for upcoming meetings and generating briefs when one is 15 minutes away.

---

Recipe 2: Action Item Tracker

Runs hourly. Extracts action items from recently completed meetings and posts them to your activity feed (or optionally creates Asana/Jira tasks).

Configuration

Agent ID: lma-action-items
Model: smart
Schedule: interval (every 60 minutes)

Tool policy:

[
  {"group": "outlook_builtin"},
  {"group": "user_mcp__live_meeting_assistant_lma"},
  {"group": "kg_tools"},
  {"effect": "allow", "tool": "live_meeting_assistant_lma__search_lma_meetings"},
  {"effect": "allow", "tool": "live_meeting_assistant_lma__get_meeting_summary"},
  {"effect": "allow", "tool": "live_meeting_assistant_lma__list_meetings"}
]

Agent prompt:

You are an action item extraction agent. Your job is to find newly completed
meetings and extract actionable commitments.

## Process

1. Call `live_meeting_assistant_lma__list_meetings` to find meetings completed
   in the last 2 hours (use startDate/endDate filters).

2. For each meeting found, call `live_meeting_assistant_lma__get_meeting_summary`
   with includeActionItems=true.

3. Extract action items. For each:
   - Identify the owner (who committed to doing it)
   - Identify the action (what they committed to)
   - Note any deadline mentioned
   - Note the source meeting (use the meetingUrl from the response so the
     user can jump to context)

4. Post a summary to the activity feed using `update_feed` with
   importance="important":
   - Group action items by owner
   - Include the meeting name, date, and meetingUrl for each
   - Flag any items assigned to the user specifically

## Guidelines
- If no new meetings are found, call skip_cycle (nothing to report).
- Focus on concrete, actionable commitments — not vague discussion points.
- Never fabricate action items — only report what's explicitly in the summary.

---

The skills pack

The amazon-quick-desktop-skills-pack/ bundle wraps the most popular workflows as installable Quick Desktop skills (one-line trigger phrases) and agents (the scheduled recipes above, pre-configured). Install the pack and you get:

• lma-meeting-catchup — “catch me up on the standup I missed this morning”. Finds the meeting from your calendar, pulls the LMA summary and transcript, optionally cross-references Slack threads from the same window, and synthesizes a personalized catch-up with deep-links.
• lma-live-coach — “coach me on this call”. Launches the LMA virtual participant into your current meeting and spawns a background agent that polls the live transcript and posts coaching cards (MEDDPICC, SPIN, Challenger) to your activity feed in real time. Inspired by KenAI Live Call Coach by Ken Beauvais.
• lma-pre-meeting-brief and lma-action-items agents — pre-configured versions of Recipes 1 and 2 above.

Skills are just packaged versions of conversational workflows. Anything in the skills pack you can also do ad-hoc with a chat prompt (see Conversational workflows) — the pack just gives you a short trigger phrase and a curated prompt.

---

Troubleshooting

”Tool not found” error

Verify the MCP connection is active:

• Settings → Capabilities → MCP → confirm LMA shows Connected
• If disconnected, click the connection and re-enter the API key

”Permission denied” on meeting data

LMA enforces UBAC — non-admin users only see their own meetings:

• Verify you’re using your own API key (not a shared one)
• For team-wide agents, ask your LMA admin for an admin-scoped key

Agent triggers but finds no meetings

• Confirm meeting history exists in the LMA web UI
• The pre-meeting agent searches based on attendee names — make sure meeting participants in LMA match your calendar invitees

Agent doesn’t trigger before meetings

• Verify the trigger config: Show me the lma-pre-meeting-brief agent config
• Check that the schedule interval (5 min) is short enough to catch the 15-min trigger window
• Confirm Outlook/Google calendar is connected in Settings → Capabilities → Connections

Write-tool prompts asking for allow_unscoped_write_tools

LMA’s read-only tools (search_lma_meetings, get_meeting_transcript, get_meeting_summary, list_meetings, get_virtual_participant_status) are annotated readOnlyHint: true and should not require this flag.

If you’re seeing the prompt, it’s because your tool policy includes a write tool (schedule_meeting or start_meeting_now). For read-only agents like the pre-meeting brief and action item tracker, exclude those two tools from the policy and the prompt should disappear.

General connection / authentication issues

See the troubleshooting section in Amazon Quick MCP Setup.

---

Reference

• Amazon Quick MCP Setup — connect LMA to Quick Suite or Quick Desktop
• amazon-quick-desktop-skills-pack/ — installable bundle of these agents and skills
• MCP Servers Overview
• MCP API Key Authentication