How to Make Claude Remember Your Preferences
Claude is sharp in every conversation. The problem is it meets you from scratch every single time. Your name, your job, how you like responses formatted, whether you want bullet points or prose — none of it carries over. Custom instructions help, but they fill up fast and are a static block of text, not a living store.
This guide shows how to use Stash — a lightweight MCP record store — to give Claude a persistent, queryable memory of your preferences, so it can pull exactly what's relevant at the start of any session.
The custom instructions problem
Custom instructions give you roughly 1,500 characters to describe yourself to Claude. That sounds like enough until you realise your role, organisation, writing style, preferred output format, timezone, current projects, and response style preferences all have to fit in that box.
They don't. Most people end up with something like:
Which is fine as far as it goes. But Claude still doesn't know which product you manage, your current sprint priorities, how you prefer to structure a PRD, or your standing constraints. And every time those change, you're back to editing a text box.
What Stash lets you do instead
Stash is a hosted MCP record store. Connect it to Claude once — it exposes context()
to fetch your standing preferences and search() to retrieve specific records on demand.
You store preferences as structured records. Claude fetches them at the start of a session. When your preferences change, you update the record — no prompt-editing required.
What to store
Five categories cover most of what matters:
- Identity: name, role, organisation, timezone
- Work context: current projects, immediate priorities, team structure
- Output style: bullet points vs prose, length, sections you always want
- Tone: formality level, phrases to avoid, UK vs US English
- Standing constraints: things Claude should never do without asking first
Setting it up
Step 1 — Add Stash to Claude. Paste the connector URL into Claude → Settings → Connectors → Add custom.
Step 2 — Create an "about-me" record. In a Claude chat with Stash connected:
Step 3 — Start future sessions with one call.
Got it — Sarah, PM at Luma, Q3 roadmap + API tier. Responses will be concise bullets. What are we working on?
One call loads your entire standing context. Claude knows who you are before you've typed the actual request.
More things worth storing
Updating preferences
When something changes — new role, new project, new output style:
Updated. Congratulations on the promotion.
Next time you call context(), it returns the updated record. No prompt editing, no copy-pasting.
Honest limitations
- No semantic search. Stash uses full-text search — keyword-based, not embedding-based. Specific terms work well; fuzzy concept matching does not.
- Not a Notion replacement. Stash is a lightweight record store, not a document editor. Use Notion for your wiki; use Stash for the context Claude actually needs.
Free tier
Stash's free tier includes 2,500 records and 50 searches per month — more than enough for personal preferences and context records. Pricing may change; current limits on the pricing page.
Get started
Paste into Claude → Settings → Connectors → Add custom. Google sign-in, free account, no credit card.
Related: How to give Claude persistent memory · The "Start my day" pattern · Why custom instructions aren't enough