How to Give Claude Persistent Memory

Claude forgets everything when you close the tab. Here's the step-by-step way to give it genuine persistent memory — across every conversation, automatically. No code, no API key, 10 minutes.

The problem with Claude's memory

Claude is stateless by design. When you close a conversation, everything in it is gone — your name, your job, what you were working on, what you asked last week. The next conversation starts cold.

Claude does give you two escape hatches, and they're both real but limited:

• Custom instructions — a ~1,500-character box that loads before every chat. Good for your name and writing style. Not good for lists, records, or anything that changes.
• Projects — a project-scoped context with a larger knowledge base. Great for a specific piece of work. Still not persistent across all your chats, and still static.

If you want Claude to remember something specific — an action item, a contact, a note, a running list — neither of those really does it. You end up copying and pasting, re-uploading CSVs, or just re-explaining yourself every time.

There's a cleaner way.

The MCP approach: a connected record store

Claude supports a standard called MCP (Model Context Protocol). It lets Claude connect to external services and use them as tools — including tools for storing and searching records.

Stash is a hosted MCP record store. You connect it once, and from that point Claude can:

• Add records — "stash this note", "save that contact", "log the outcome"
• Search records — "what did I save about Acme?", "pull my top tasks"
• Load standing context — role, org, current OKRs, writing style — into every fresh conversation

The memory lives in Stash, not in Claude. So it persists across every conversation, every device, every session.

What you can store

Stash uses collections — think named lists. You can have as many as you need. Common ones people set up in the first session:

• context — who you are, what you're working on, your writing style. Loaded into every chat automatically.
• tasks — open action items. "What's on my list?" gets them all back, instantly.
• contacts — people, deals, follow-ups. Your pocket CRM.
• notes — anything you want to recall later. Searched with full-text search (FTS5).
• reads — articles and summaries to come back to.

Everything is searchable. Everything is yours — export or delete anytime.

Step by step: setting it up

1. Copy the connector URL

   Stash is a hosted service — nothing to install. Copy the URL below, which is the MCP endpoint Claude will connect to.

2. Open Claude → Settings → Connectors

   In Claude.ai, go to your profile icon → Settings → Connectors. Click "Add custom" (or "Add integration" — the label varies).

3. Paste the URL and connect

   Paste the connector URL. Claude will ask you to sign in with Google — this creates your free Stash account and ties the store to your Google identity. No credit card needed.

4. Add one line to your custom instructions

   Go to Settings → Custom instructions. Add: At the start of each conversation, call context() from my Stash to load my standing context.

   This is optional but powerful — it makes every new chat load your standing context automatically.

5. In your first conversation: set up context

   Tell Claude your role, what you're working on, your writing style, your current priorities. Ask it to save this to your Stash context collection. From then on, every new conversation starts knowing who you are.

What "Start my day" looks like in practice

With the custom instruction in place, here's what the first message of a new conversation looks like under the hood:

▸ loading context…
· role: product lead, Series A SaaS
· org: Acme (18 people, 3 squads)
· writing style: direct, no filler, bullet-heavy
· top tasks → tasks collection
· current OKR: land 5 enterprise pilots by Q3
▸ ready · one cheap call

No re-explaining. No copy-pasting. Claude just knows. That's persistent context that actually works.

The honest tradeoffs

For most use cases — lists, contacts, tasks, quick notes, standing context — FTS5 is exactly what you want. It's faster, cheaper per query, and never hits a row wall. The semantic search gap matters for research retrieval; for everything else, keyword search wins.

Free tier limits

The free tier is genuinely usable: 2,500 records and 50 queries per month. That's plenty for a standing context, a task list, a CRM, and a reading list. If you're running heavier workloads, Pro is £8/month and gives you 100,000 records and 1,000 queries. No credit card for the free tier.

Pricing may change — this is a new service. If it does, you'll get 30 days' notice. Cancel anytime, no lock-in.

Ready to add it?

Ten minutes to set up. Your Claude will remember things from the first session forward.