Senses · 02: Connectors

Memory tells the system what your business knows. Connectors give it hands in the tools you already use. Without this layer, every task is copy-paste and you’re the go-between. Connectors are what let the Cortex read an email, file a note, and update a record without you in the middle.

Why connectors are the second sense

Claude Code is the surface you type into. Connectors are how it reaches everything you don’t type into: your inbox, your calendar, your notes, your automations. Claude can write a great reply, but if it can’t read the thread or draft the response in your email, you’re still the courier carrying data back and forth.

There are two kinds of connector, and you want both:

A connector that lets Claude use a tool live, right in the session. Ask it to search your email or pull up a file, and it does, in the conversation, with you watching. The technical name is MCP. You don’t need the acronym, just the idea.
A tool that runs automations on a schedule, with no one watching. This is n8n. It’s where a task like “every morning, sort my inbox” runs on its own. The Habits chapters build the actual automations.

The first is for “help me with this now.” The second is for “do this every morning without me.” Together they’re the Cortex’s full set of eyes, ears, and hands.

One thing before you wire anything: this is the layer with real reach. A connector that can send email or delete records is real power handed to software. The discipline matters more than the wiring. Give each connector the least access that does its job, and keep a written map of what’s connected.

The install

1. Live connectors (help me now)

Wire the tools you actually work in. Start with the few you touch daily. Add the rest when a real task needs one.

Connector	What Claude can do with it	Typical use
Email	Search, read, draft, label	Sort the inbox, draft replies in your voice
Files (Drive)	Search, read, create	Pull a doc, file a new one
Calendar	Read and manage events	Prep, scheduling context
Notes (Notion)	Search, read, create, update	Read and write your knowledge base

The rule: wire a connector when a task needs it, not because you can. Each one is access you granted and a thing that can break.

2. The automation tool (do it without me)

Stand up n8n, the tool that runs your automations on a schedule. Here you just make sure it exists and Claude can reach it. The Habits chapters build the actual flows. Think of it as the engine; the workflows come later.

3. Secrets, kept outside the vault

Passwords, keys, and tokens go in a password manager or a locked settings file, never in a note and never in your Codex. Your Codex is plain text and it travels: you sync it, you might share it, it lives in version history. A secret written there is a secret leaked. Keep them out.

4. The connectors map

One small doc that’s the single source of truth for what’s wired:

# Connectors Map

## Wired (live)
| Tool | Access | Notes |
|---|---|---|
| Email | read + draft, NOT auto-send | drafts only, I hit send |
| Notes | read + write | my workspace only |
| n8n | build + run workflows | runs on a schedule |

## Deliberately NOT wired
| Tool | Why not |
|---|---|
| Bank / payments | Never. No automated access to money. |
| Deleting customer records | Read and add only. Deletes stay manual. |

The “Deliberately NOT wired” section is the most important part. It records what the Cortex must never touch. That’s a decision worth writing down, not a to-do list.

Pitfalls (the “if this, then that” guide)

Situation	What to do
You wire every connector you can find	Don’t. Each one is access plus upkeep. Wire the daily few, add others on demand. An unused connector is risk for no benefit.
You give a connector broad access “to be safe”	Backwards. Broad access is the risk. Read-only where you can, draft-not-send for email, no delete on records that matter. Least access by default.
A password lands in a note or your Codex	Pull it now, change the key, move it to the password manager. Your vault travels, so a secret in it is a secret out.
You let email send on its own	Don’t, at least not early. Draft-only with you hitting send is the safe default. Earn auto-send over time, if ever.
You can’t remember what’s connected	That’s what the map is for. If it doesn’t exist, you’ve lost track of your own reach. Write it.
You think about wiring your bank or payments	Never. There’s no version of “the system moves money on its own” that’s worth it. Money movement stays human, full stop.
A connector breaks quietly and a scheduled task fails	Add a health check. A connector that can fail silently will, and an unattended task won’t warn you unless you set up the alert.
Your info is scattered and capturing it feels impossible	Real problem. Don’t capture it all at once. Start by hand, automate later, and wire meeting capture so calls land on their own. The Cortex grows from the trickle of what you flag, not from moving everything on day one.

Maya’s version

Maya wired four connectors, in her business shape, and stopped there. Email, so Claude can sort her inbox and draft supplier replies in her voice. Her files, so it can pull a product photo or a past invoice. Her calendar, for scheduling. Her notes, where her Codex and customer info live.

Draft-not-send is on for email from day one. Claude writes the reply; she reads it and sends it.

Her store runs on a payment processor. She never wired it, and she wrote that in the “Deliberately NOT wired” row: no automated access to money, ever. Her keys sit in her password manager, not in her notes.

Two things worth keeping:

The map is shaped by what it won’t do as much as what it will. Anyone can see, in one doc, exactly what her system can and can’t touch. That’s what makes it safe to run.
Live access and scheduled automation are two different jobs. One helps her in the moment. The other runs while she sleeps. She set up the first now and left room for the second.

You’ve got it when

Your daily tools (email, files, calendar, notes) are wired and usable inside a session
The automation tool (n8n) is running and Claude can reach it
Every key and password lives in a password manager or locked settings file, none in the vault
A connectors map exists, including the “deliberately NOT wired” decisions
At least one connector is read-only or draft-only on purpose
You can ask Claude to read from one tool and write to another in one session, and answer “what is the Cortex allowed to touch?” from a single doc

If the tools are wired and you can point to the map of what they’re allowed to touch, the sense is live. The last two prove it’s governed, not just connected.

The skills that help here

The automations are built with the /n8n-* commands (like /n8n-build and /n8n-fix) in the Habits chapters; the live connectors are set up in Claude Code’s settings, not as slash commands. Setup notes are in the starter kit and ../03-habits/01-skills-and-commands.md.

→ Next: 03-the-interface.md. The Cortex has no single front door. Learn which surface fits which job: build in the terminal, capture by voice, review on a dashboard.