Working with AI assistants
Prompts and workflows that get the most out of Memo AI inside Claude, ChatGPT, Cursor and Claude Code.
Once the MCP server is connected, you talk to your meetings in plain language. This guide covers prompts that work well and the patterns behind them.
Good first prompts
| Goal | Try saying |
|---|---|
| Find a meeting | "Find my meetings with Acme from last week." |
| Summarise | "Summarise yesterday's standup and list action items." |
| Compare | "What changed between my last two calls with the design team?" |
| Extract | "Pull every decision we made about pricing this month." |
| Export | "Save this week's meetings to ~/notes/meetings as markdown." |
Why mention "Memo AI"?
Assistants decide when to use tools from your wording. Naming the source — "in Memo AI", "my meetings", "my transcripts" — reliably triggers the right tool. Vague prompts may get answered from the model's general knowledge instead.
Multi-step workflows
The assistant chains tools on its own. For "summarise my Acme calls this month and save them" it will typically:
memo_list_transcriptions— filter by date, find the Acme meetingsmemo_get_transcription— read each one to write the summarymemo_bulk_export— save them all to your folder
You don't orchestrate this — just describe the end result.
Tips
- Be specific about dates and names. "Last week", "the Acme call", "the Q2 planning meeting" all help the assistant filter precisely.
- For exports, give a path. "Save to
~/notes" lets the assistant write files directly withoutput_path/output_dirinstead of dumping text into the chat. - Long meetings → ask for export, not paste. Exporting to disk avoids loading a huge transcript into the conversation.
- Stay in your workspace. A key sees one workspace; if a meeting seems missing, you may be on a key for a different workspace.
Which clients work?
Any MCP-compatible client: Claude Desktop, Claude Code, Cursor, ChatGPT (Plus/Pro/ Team connectors), and others. Setup for each is in the MCP Quickstart.