MCP recipes

Setup done (MCP setup)? These are the patterns that make it pay off. They share one idea: chat AIs are good at one-off answers, but a connected AI can do sustained work on your actual workspace.

Write it to Banter

The habit that changes how you use AI: end research and drafting sessions with the output filed, not stranded in a chat.

…great. Now write that up as a note in /Research called “Vendor comparison — CRM options”, short bullets, include the pricing table.

Chat history rots — nobody rereads last month’s threads. A note in the right folder keeps working: it shows up in search, the assistant cites it, and next quarter’s you finds it. Do this with everything worth keeping.

Bulk import

Covered in detail in Import your notes: export your old app as markdown, then have Claude Code propose a structure and file everything. This is the flagship bulk operation — hundreds of notes filed in one session, something no chat UI or manual process does well.

Workspace cleanup

The reorganization you keep postponing is a good agent task:

Look at /Inbox in Banter. Propose where each note should be filed based on my folder structure — then move them once I confirm the plan.

Find notes in /Work with useless titles like “Notes” or “Call” and rename them based on their contents.

Ask for the plan first, then the execution. Agents are diligent movers and honest about ambiguity when you make the confirm step explicit.

Meeting prep, from anywhere

Your workspace answers questions wherever your AI runs — including clients that also see your calendar or email:

Before my 2pm with Acme: read the Acme folder and the last two call transcripts in Banter, and give me open items, promises we made, and what changed since last month.

Standing instructions make it compound

Every rule you put in your workspace instructions applies to connected agents too — filing conventions, writing style, what to always include. The more your workspace teaches, the less each prompt has to.