The problem isn't taking meeting notes. Everyone does that. The problem is what happens next: the notes sit in a doc somewhere, the action items slip out of memory, and three weeks later someone asks "what did we decide about X?" and nobody knows where to look.
Claude is very good at processing notes, pulling out action items, and drafting follow-ups. But Claude forgets everything when the conversation ends. The work you do together is gone unless you store it somewhere it can be recalled.
This post covers a simple pattern: using Stash (a token-light MCP record store) to keep your meeting notes searchable and surfaceable to Claude in future conversations. No new apps. No complex setup. The connector adds to Claude in 30 seconds.
In a fresh conversation, you can hand Claude a transcript or rough notes and it will do excellent work: extract action items, group by owner, draft follow-up messages, flag blockers. You get a clean output in minutes.
But the next day, you're starting fresh. Claude has no idea who Sarah is, what project you discussed, or what was decided last Tuesday. Every new conversation means re-establishing context.
For one-off meetings this is fine. For recurring work — weekly project standups, client check-ins, team planning sessions — the friction compounds. You end up spending the first part of every Claude session re-pasting last week's output before you can get today's work done.
Stash keeps records in a full-text searchable store that Claude can query over MCP. Instead of pasting context into every session, you write it once and retrieve it with a search.
The pattern looks like this:
The cost to retrieve: roughly 200–400 tokens for a targeted search. Not 4,000 tokens of raw notes pasted in full.
Here's an example with a recurring project standup:
You: Here are my notes from today's product standup: - Sarah to write the API spec by Friday - Awaiting legal sign-off on the privacy clause before we can launch feature X - Tom flagged: the onboarding modal is broken on mobile, needs fix before 1.8 - Decision: we're keeping the current pricing tier; revisit in Q3 Store these as action items in my "product" collection in Stash. Claude: Done. I've added 4 records to your "product" collection: - [action] Sarah: API spec by Friday - [blocked] Feature X: waiting on legal / privacy clause - [bug] Tom: onboarding modal mobile — target 1.8 - [decision] Pricing: hold current tiers, revisit Q3
Next week, before the standup:
You: What's still open in the product collection from last week? Claude: Searching Stash for open items in "product"... Found 3 open items: - Sarah's API spec (due Friday — check status) - Feature X still blocked on legal (no update since Thursday) - Mobile onboarding bug — was it fixed before 1.8? One item closed: pricing decision is marked done.
Two messages. Zero re-pasting. Claude is fully in context.
Stash lets you group records into named collections. For meeting notes, a simple structure works well:
| Collection | What goes in it |
|---|---|
product | product decisions, feature flags, design choices |
clients/acme | per-client action items and decisions |
hiring | candidate notes, interview decisions, offer status |
weekly | rolling standup and weekly review notes |
decisions | cross-project architectural or strategic decisions |
You can search across all collections (search("onboarding")) or restrict to one (search("onboarding", collection="product")). Full-text search, so "API spec" finds "the spec for the authentication API" too.
One useful pattern is a simple status field in each record. You can tell Claude to update the record when something closes:
You: Sarah's API spec is done. Mark that action closed. Claude: Updated the record in your "product" collection. Status changed to closed. I've also added a note: "completed 2026-06-07".
And when you ask for what's open next week, closed items don't come back unless you ask for them.
This isn't a project management tool — it's deliberately lightweight. It stores records, searches them, and surfaces them to Claude. For complex task management you'd use Jira or Linear. Stash is the layer that keeps Claude informed about what those systems contain, without requiring Claude to log into anything.
Once you have meeting notes and action items in Stash, they plug into the Start-my-day workflow naturally:
You: Good morning. What's on for today? Claude: [calls context() then searches Stash for open items] Here's what I see: - Your standing context: product lead at Acme, timezone GMT, current sprint ends Friday - 2 open blockers from last week's standups - Sarah's spec was closed yesterday — might be worth a quick review before standup - 1 unread decision record from Thursday's architecture call
This is what "Claude as a secretary who actually remembers" looks like. The context isn't pasted — it's searched, filtered, and surfaced. Claude pulls what's relevant for today, not everything you've ever stored.
Stash is a hosted MCP connector — you add it to Claude and it's ready.
No installation. No server to manage. A free account gives you 10,000 records and 100 searches per month — enough for months of meeting notes before you'd think about upgrading.
It's not a meeting recorder, a transcript service, or a real-time notes app. You're still taking the notes yourself (or using a dedicated tool for that). Stash is the retrieval layer — where the structured output from your notes lives so Claude can access it across sessions.
It also won't replace a proper task manager for complex projects. It's the layer between "too small for custom instructions" and "too heavy to paste in" — specifically optimised for getting the right things back to Claude cheaply and quickly.
Keep meeting notes in Claude's reach
Add Stash to Claude. Free to start. No installation required.
Add to Claude →