
Agents You can start this yourself today.
Let an agent read your brand without letting it loose.
Claude, ChatGPT or Cursor gets one brand at a time. It reads a bounded part of your record, and any change it suggests waits for you.
At a Glance
Point your AI agent at your brand.
- What connects
AI tools that speak MCP (the open protocol they plug into). We have tested Claude, Claude Code, ChatGPT, Codex and Cursor: each signs in with Jinn or carries one token, and reads your record instead of inventing its own version of your brand. Add Jinn to each assistant you use. The ones you sign in on share one brand record.
- What a demo token reads
A curated part of a showcase brand's record: its identity, its voice rules including the words it never uses, and its positioning. A demo token can only read.
- What stays out
On a demo or Connected plan, your competitive notes, your pricing and your product internals stay out. They are left off the public part of the record, and tests check that they stay off. Only your own Brand or Agency plan reads the full record.
- What you can try today
The demo. One request makes a short-lived token for a few showcase brands, with no account and no card.
Any AI agent signs in with Jinn and reads your brand's record. Start in the assistant you already use: sign in with Jinn there, let it read your brand record, and any change it suggests waits for your approval.
One connection, one bounded read

Its limits, read first
Plug checks the key.
Before it reads anything, the agent asks which brands it may read and on which plan.

A bounded read
Ocho opens one binder.
It reads only what its plan allows. A brand it was never given reads back as if it did not exist.

A suggestion that waits
Drift files the slip.
A change it asks for becomes a proposal in your inbox, and nothing in the record moves until you say yes.
The connection
One way in, a bounded read out.
The server checks the boundary on every call; it does not trust whatever a client sends.
- Sign in or request
Sign in with Jinn from a tool that supports it. Free to try: one brand, with a preview of your brand record. The full record starts with the Brand plan. A token, made in your account, is for scripts and other MCP clients. For a demo there is no signup and no card: one request returns a short-lived token for showcase brands.
- Check its limits
Before it reads anything, the agent calls one tool to see its own permissions. It learns its limits from the token instead of probing to find them.
- Read
It reads what its plan allows: on every paid plan, the public part of the record and the design system as a document and as tokens; from Connected up, answers to plain questions about the brand; from Brand up, the full record, the kit and the products. A free sign-in reads your brand's status.
- Stay in bounds
The server checks every call against the token's brands and plan, never against what the client claims.
The Connection Map
From your agent to your brand record.
Claude, ChatGPT, Cursor or another AI tool that speaks MCP.
Sign in with Jinn where your tool supports it. A token is for scripts and other MCP clients.
Every call is checked against the token's brands and plan, never against what the client claims.
Each call is logged: the tool, the brand, the time and a fingerprint of the request, never the request itself.
The agent reads only what its plan allows, and any change it suggests waits for your approval.
Honest edges
What a paid plan adds, and what it doesn't.
"We make it easy to build from your brand record."
| Plan | What the agent can reach |
|---|---|
| Demo token | The public part of the showcase brands' records, and their design system as a document and as tokens. It can only read. |
| Connected | Your own brand, with no expiry. Adds answers from your record, a brief before a task, a search and a review of your own assets, your brand as a skill folder, and proposed edits that wait for you. |
| Brand | Everything in Connected, plus the full record, the kit, the products and the design system measured from your site. |
A call to a tool above your plan returns a clear error with the upgrade path, never a silent failure. Agency pricing is on the plans page.
The boundary
Bounded on purpose, and checked by tests.
The telling part of a public brand surface is what it refuses to show. Here the refusal is built in.
A curated part, not the record
The public part of the record is an explicit list of fields: identity, voice, positioning. Everything else is left out by not being on the list, and a field renamed upstream fails the build instead of leaking.
The valuable parts are left out
Competitive notes, pricing and product internals are not in the public part, and tests check that they are absent. They are not hidden behind a switch; they were never put in.
A token can't widen itself
Every self-serve token has its audience and scopes fixed on the server when it is made, whatever a client or route sends. There is no scope for a client to forge.
An ungranted brand looks like nothing
Asking for a brand outside the token's list returns exactly what a brand that doesn't exist returns, so a token can't even confirm which brands it wasn't given.
Changes
An agent can suggest. Only you can change.
On a paid plan, an agent can propose an edit to named fields of your record. Each proposal lands in your learning inbox, and the record stays put until you approve it.
No tool on the connection posts or publishes anything.
Signed in with Jinn, an agent can also switch which of your brands it reads. The switch applies from its next call.
Honest machinery
Bounded by the server, not by trust.
A demo needs no account
The demo path is real and open today: one request makes a short-lived token for the showcase brands, with no signup and no card. You can point an agent at it now.
The record underneath
What an agent reads is drawn from the same verified brand record behind every Jinn feature: 700+ signals from your own site and evidence. Agents read the record. They don't invent one.

Agents: want to see it run on your brand? Let's talk.
- Tell us about your brand.
- A person on our team reads it.
- We read every application and reply within two business days.
