Search Your Notes with Claude via MCP

June 2026 · 6 min read

Searching notes is a solved problem. You open Notion, Bear, Obsidian, or Apple Notes and type into a search box. But there's a different kind of note retrieval that's becoming useful: asking Claude to find something from your notes in the middle of a conversation.

"What did I write about the Acme deal last month?" "What was that thing I read about LLM context management?" "Find my notes on the Harris framework." In natural conversation, without switching apps.

MCP makes this possible. With a searchable note store connected to Claude, you just ask.

How this works

The Model Context Protocol (MCP) lets Claude connect to external data sources over HTTPS. When you add an MCP server to Claude, it gains new tools — in Stash's case, tools to add records, search them, and retrieve results.

Once connected, Claude can call those tools mid-conversation without you asking explicitly. Say "find my notes about X" and Claude searches your Stash, reads the results, and incorporates them into its reply.

A real example

You: I'm writing a proposal for a client. Can you find any notes I have on their previous feedback?

Claude: [searches stash for "client feedback proposal"]

Claude: Found three notes tagged "client-feedback": one from your March call (they flagged pricing transparency), one from the April demo (they loved the reporting dashboard, asked about API access), and a note from last week where you wrote "ask about their integration timeline." Want me to draft the proposal with those points in mind?

You: What's in my "reading" collection?

Claude: [retrieves reading collection — 12 records]

Claude: 12 items. Most recent: "Attention Is All You Need" (tagged: ML, papers), "The Lean Startup" (tagged: books, strategy), "MCP deep-dive by Simon Willison" (tagged: mcp, ai-tools). Want me to pull any of those up in full?

You: Add a note — "talked to Sarah, she's interested in the enterprise tier, follow up in two weeks"

Claude: Added to your notes. Tagged "contacts, follow-up". I'll surface this if you search for Sarah or follow-ups.

Setting up Claude as a note-search assistant

Sign up for Stash at stashlite.com (Google OAuth, takes 20 seconds). You'll get a connector URL.
Add the connector to Claude: Settings → Connectors → Add custom MCP server → paste your URL → Connect.
Create your first collection:
In Claude: "Create a Stash collection called 'notes'"
Start adding notes in natural language during conversations:
stash it — "key insight from today's meeting: ..."
Search anytime:
search stash for notes about [topic]

What kinds of notes work well

Stash stores records — short to medium-length text with a title, tags, and a body. It's not a document editor (no rich text, no image embeds). What it's good at:

Meeting notes — who said what, what was decided, what needs follow-up
Reading highlights — quotes and annotations from things you've read
Contact notes — context about people before calls
Project context — the "state of the world" for ongoing work
Quick ideas — capture-first, organise later

Token-light by design:
Stash search results return titles, tags, and a short excerpt — not the full body unless you ask. A 500-record search costs ~192 tokens (measured, preliminary). Claude's context window stays lean even with a large note collection.

What it doesn't do

Stash is not a replacement for a notes app like Obsidian or Bear for long-form writing, bidirectional linking, or rich formatting. If you write detailed documents with headers, images, and internal links, you still want a dedicated notes app for that.

Stash is the layer that Claude can search. Think of it as your Claude-native note index — the records Claude actually reaches for when you ask it something.

Start in 30 seconds

Sign in with Google →
Get your connector URL. Add to Claude. Call context() to confirm it works. Free tier includes 10,000 records and 100 search queries per month.