Claude is genuinely excellent for research. It can synthesise papers, generate hypotheses, find connections between ideas, suggest angles you hadn't considered, and help you think through what you're actually trying to learn.
The problem is structural: every conversation starts blank. Whatever you found in Monday's session — the sources you evaluated, the contradictions you noted, the thread you decided to follow — is gone by Tuesday. If you're doing research that spans more than one conversation, you're constantly re-explaining your own project to yourself.
This post shows a concrete pattern for fixing that. It takes about 30 seconds to set up and changes how multi-session research with Claude works.
Claude's context window is large, but it resets completely at the end of a conversation. There's no background sync, no notebook, no place it writes things down between sessions. This is fine for one-shot questions. It's genuinely painful for research projects that might span weeks.
The workarounds most people try:
None of these solve the underlying problem: you want Claude to know what you've already found the moment a new conversation starts, and you want to search your accumulated research by topic, keyword, or tag without pasting anything.
Stash is a small knowledge store that connects to Claude via an MCP connector. When you add the connector, Claude can save notes directly to your personal store during a session. Next session — or next week — it can retrieve them.
The research pattern looks like this:
No copy-pasting. No uploading docs. The notes are already there when you need them.
Say you're researching the competitive landscape for a product you're building. Session 1:
Session 2, a week later:
Claude has full access to what you've already accumulated. The conversation starts where the last one ended, not from zero.
You don't have to wait until the end of a session. The most useful pattern is saving as you go — whenever you find something worth keeping:
Low friction. One line. The note is there when you need it.
When you have dozens of notes across several collections, search becomes the interface:
FTS5 full-text search on every field. Search by keyword, tag, or collection. The retrieval is fast and token-light — you're not loading your entire research corpus into context, just the parts you asked for.
| Research type | What to save | How to retrieve |
|---|---|---|
| Competitive analysis | Player summaries, pricing notes, positioning observations | Search by competitor name or tag |
| Academic / literature review | Paper summaries, key claims, contradictions | Search by topic, author, or methodology tag |
| User interviews | Interview notes, recurring themes, direct quotes | "What have interviewees said about X?" |
| Market discovery | Data points, sources, hypotheses tested | Search by hypothesis or date range |
| Due diligence | Findings, red flags, open questions | Pull all open questions, all red flags |
The biggest time-saver is the context() tool. You put a short standing brief in your Stash — what project you're working on, what you're trying to find out, where you've got to — and call it at the start of a fresh conversation:
One tool call. 15 seconds. Claude knows your project, your progress, and your next move. You can skip the preamble and go straight to work.
context() returns your standing brief — typically 50–200 tokens. Searching a collection returns only the matching records. Compare this to uploading a research doc at the start of every session, which typically costs 2,000–20,000 tokens before the conversation even starts. The pull model is cheaper and more targeted.
In Claude: Settings → Connectors → Add custom, then paste:
https://app.stashlite.com/mcp
Sign in with Google. That creates your account. Then tell Claude what your research project is and ask it to save a standing brief:
I'm researching [your topic]. My goal is [goal]. Save this as my standing context so I can start future sessions with context().
You're done. Every session from here can start with context() and build on everything you've saved.
Stash is free up to 2,500 records and 50 searches per month — enough for most research projects while you're getting started. Pro is £8/month with effectively unlimited records and searches. Pricing may change; cancel anytime.
If your research generates more than 2,500 notes, you probably need the Pro tier — but 2,500 is a lot of research notes.
Persistent notes. Instant retrieval. Sessions that start from where you left off.
https://app.stashlite.com/mcp