June 2026 · 6 min read

How to Save Information in Claude (So It Remembers Across Sessions)

You just had a great conversation with Claude. It helped you draft a strategy, remembered your company context, got the tone exactly right. You close the tab.

Next session: blank slate.

Claude has no memory of any of it. Every conversation starts from zero. If you work with Claude daily, this gets old fast — especially when you're constantly re-pasting the same background context or re-explaining who you are and what you're building.

There are four ways to fix this. Here's an honest breakdown of each.

Option 1: Custom Instructions

Claude.ai's custom instructions let you set a short block of text that gets prepended to every conversation automatically. You write it once; Claude reads it every time you open a new chat.

This works well for:

• Your role and communication style ("I'm a product manager, answer concisely")
• Standing preferences ("always use British English", "prefer tables over prose")
• A brief professional bio

Good for: standing preferences, short bio. Not good for: lists, dynamic context, or anything over ~200 words.

Option 2: Claude Projects

Projects (available on Pro and Team plans) let you create a persistent workspace with uploaded files and a project-level prompt. Claude keeps context within a project, and you can upload PDFs, text files, or docs as reference material.

This works well for:

• Ongoing work with a defined scope (a client project, a writing series)
• Reference docs you want Claude to be able to cite
• Team-level knowledge bases (Team plan)

Good for: scoped reference material. Not good for: dynamic lists, programmatic access, or data that updates regularly.

Option 3: Paste It Every Time

This is what most people actually do. You keep a text file (or a Notion page, or an Obsidian note) with your standing context — your role, current projects, contacts — and paste the relevant section at the start of each conversation.

It works. It's also annoying, error-prone, and uses a lot of your context window before you've even started.

For a 500-record contact list, pasting the full list into every conversation consumes roughly 50,000–100,000 tokens per session (at current Notion export verbosity). That's expensive and slow.

Good for: quick one-off contexts. Not good for: anything you do regularly, or data with more than a few hundred items.

Option 4: An MCP Record Store (the right tool for lists and dynamic context)

Model Context Protocol (MCP) connectors let Claude talk directly to external services — search them, read from them, write to them — during a conversation, without you pasting anything.

An MCP record store is a hosted service that:

• Stores your lists (contacts, tasks, notes, Notion exports) outside Claude
• Lets Claude retrieve exactly what it needs via a search query — not the full dump
• Keeps a standing context block that loads automatically when you call context()
• Updates dynamically: add a record in one conversation, it's there in the next

The token cost difference is significant. Searching a 500-record store via an MCP tool call costs around 150–200 tokens (the query + structured result). Pasting the same list directly costs 50x–100x more.

Which option is right for you?

In practice, most heavy Claude users use all four: custom instructions for standing preferences, Projects for scoped work, and an MCP store for the dynamic stuff (contact lists, task backlogs, ongoing context). They're not mutually exclusive — they solve different problems.

How the MCP approach works in practice

Here's what the Stash workflow looks like after you set it up:

First-time setup (2 minutes):

1. Sign up at stashlite.com with Google (free, no card)
2. Copy your connector URL
3. In Claude: Settings → Connectors → Add custom → paste the URL

That's it. Claude can now talk to your Stash store.

Every day after that:

You: Start my day

Claude: [calls context() — loads your standing context]
Claude: Good morning. You're mid-sprint on the API redesign
        (deadline Friday), two client calls today at 11 and 3,
        and you flagged "follow up with Sarah re: contract" last
        Tuesday. Where do you want to start?

No pasting. No re-explaining. Claude loaded your context from the store automatically — a single tool call, ~180 tokens.

Adding to your store mid-conversation:

You: Stash this: met Jamie Chen at TechConf,
     CTO at Foundry Labs, interested in our API.

Claude: Added to contacts. Jamie Chen, CTO Foundry Labs,
        TechConf 2026-06-06, interested in API.

It's there in the next session. And the session after that.

What Stash is not

To be clear about the scope:

• It's not a replacement for Notion if you need rich documents, tables with relations, or team wikis
• It's not a database if you need relational joins, aggregates, or schemas with foreign keys
• It's not long-term memory that Claude builds automatically (you tell it what to store)

It's the lightweight layer between "too small to bother" (custom instructions) and "too heavy for this use case" (Notion). Fast retrieval. Cheap token cost. Grows with you.

Further reading

• Start my day: loading standing context with context()
• Using Claude as a mini-CRM with Stash
• Offloading your Notion database to Stash
• Claude Projects vs MCP connectors: what's the difference?