Stash / Blog · June 2026 · 5 min read

Start My Day with Claude: Load Your Context in One Tool Call

Every new Claude conversation starts blank. Claude doesn't know who you are, what you're working on, or how you like things written. So most people solve it the same way: they paste their whole setup at the top of every session.

Something like: "I'm a product manager at a B2B SaaS company, currently working on a pricing redesign. I prefer bullet points over prose, brief over thorough, and I want honest pushback when I'm wrong."

It works. It also burns 80–200 tokens before you've done anything — and takes time to type or find each morning.

There's a faster way.

The context() tool

Stash is an MCP record store — a connector you add to Claude once, then use to store and retrieve structured information. One of its built-in tools is context(): a reserved collection called context that Claude loads at the start of a conversation.

The workflow:

  1. You add your context once: add(collection="context", key="me", text="I'm a...")
  2. At the start of any new conversation, you (or Claude's custom instructions) call context()
  3. Claude gets your standing context back — typically 10–50 tokens — and proceeds knowing who you are

You set it once. It persists. Every session starts oriented.

What it looks like in practice

Step 1: Store your context (do this once)

In a Claude conversation with Stash connected:

You: Add this to my context collection, key "me":
"Senior engineer at a fintech startup. Currently building a new payments
 API. Prefer short answers — give me the code, skip the explanation unless
 I ask. Call out security issues explicitly."

Claude: [calls add(collection="context", key="me", text="...")]
Done. I've stored your context. It'll load automatically next session.

Step 2: Start tomorrow's session

Add this to your Claude custom instructions (one time, in Claude settings):

At the start of every conversation, call context() to load my standing context.

Now every new conversation opens with:

Claude: [calls context()]
→ Returns: {me: "Senior engineer at a fintech startup. Currently building
  a new payments API. Prefer short answers — give me the code, skip the
  explanation unless I ask. Call out security issues explicitly."}

I've loaded your context. You're a senior fintech engineer working on
a payments API. Short answers, code-first. Ready — what are we doing today?

That's it. No pasting. Claude knows who you are and what you want before you've typed a word.

Token cost: pasting vs. fetching

MethodTokens per sessionManual work
Paste context note manually 80–250 Every session
Claude Projects (system prompt) 80–250 (always loaded) Set once, but always paid
Stash context() ~10–40 Set once, fetched on demand

The difference matters for Claude API users paying per token. If you run 20 conversations a day and each one opens with a 150-token context paste, that's 3,000 tokens/day in overhead — just for setup.

With context(), the same setup costs roughly 400 tokens/day.

Note for Claude.ai subscribers: You're paying a flat rate, not per token — so the token savings don't show up as a bill reduction. But you still benefit: less manual pasting, more conversation budget for actual work.

Expanding context: multiple collections

Context doesn't have to be a single note. You can store your context in named sections and expand into related collections on demand:

context(expand=["work", "clients", "writing-style"])

This pulls your core context plus any linked collections — so a morning where you're preparing client proposals might load:

While a quick coding session might just call context() — no expansion, just the essentials.

What people store in their context

There's no required format. Common patterns:

Store whatever you'd otherwise type at the top of every session.

Keeping it current

Context is a living document. When your situation changes, update it:

Update my context key "me" — I've moved from fintech to a healthtech startup,
 and we're now building a patient data platform instead of payments.

upsert() writes by key — so updating "me" replaces the old entry, not adds another.

Good practice: review your context once a week and update what's stale. A context that's three months out of date is worse than no context at all.

The "start my day" briefing

For people who use Claude as a genuine daily tool — not just for occasional tasks — the most useful pattern is a morning briefing. Claude opens, loads context, and gives you a short situational summary before you've asked anything.

Add this to your custom instructions:

At the start of every new conversation:
1. Call context() to load my standing context
2. If it's a fresh session (no prior messages), briefly summarise what I should
   focus on today based on my stored context and any open tasks in my "tasks"
   collection.

This turns every new Claude conversation into a brief: your role, your priorities, and your open items — loaded from Stash in about three tool calls, before you've typed a word.

Set it once. Start every day oriented.

Free account includes the context() tool and 10,000 records. No card required.

Add Stash to Claude →

Further reading