Run Personyze from Claude, ChatGPT
or any AI assistant
Connect your account over the open Model Context Protocol. Ask how campaigns are doing, find where you are losing traffic, have it draft a campaign, manage audiences — in plain conversation, with no exports and no code.
You choose what it can reach.
Not your API key — its own grant, with only the scopes you tick.
OAuth 2.1 + PKCEno key pasted into a chat window, everIts own grant, not your API key — and you can end it without touching anything else.
read, scopedthe same figures the panel shows, not a screenshot of themIt reads the account, not a scraped screen — so the answer is the number, not an approximation of it.
It built it. You still ship it.
Everything it creates arrives as a draft, switched off, waiting for you.
create_campaign_draftit cannot switch anything on by itselfDrafts, never live campaigns — the assistant does the typing, and you keep the decision.
Each step is a tool it had to be granted.
Nothing happens that you did not allow, and every call is on the record.
list_audiences14 returnedcreate_audience“Trial ending, no card”duplicate_campaigndraft #2291write, scopedthe same objects the REST API exposesTools, not a free hand — it can only do the things you ticked when you connected it.
One click, and it is out.
Each assistant has its own grant, so ending one leaves the others alone — and your API key untouched.
revokeaccess stops on the next call, not at the end of a sessionRevocable on its own — which is the reason it never needed your API key in the first place.
What it is
Your account, inside the assistant
MCP (Model Context Protocol) is the open standard AI assistants use to work with outside services — introduced by Anthropic in late 2024 and adopted across every major assistant since. Personyze runs an MCP server, so an assistant can reach your account directly instead of you copying data between tabs.
It runs on the same external API as the full-featured API, under the same permissions. Every connected assistant rides on a key belonging to your account, and can see and change only what that account can.
What you can ask
Four things people ask on day one
The wording is yours — these are kinds of question, not fixed commands.
01
How are campaigns performing?
Per-action results over any date range — impressions, clicks, dismissals, your configured goals and revenue totals.
“Which campaigns performed best in the last 30 days?”02
Where am I losing traffic?
Site-wide traffic by page, landing page, referrer, search term, device, city or country — with the personalization impressions and clicks on those pages.
“Which landing pages lose the most sessions?”03
Build me something
Describe targeting in plain words and it compiles real rules, creating the campaign in staging so live traffic is untouched.
“Create a campaign targeting returning mobile visitors.”04
Manage audiences and profiles
Create audiences, add or remove members, search visitor profiles by Personyze id, your own CRM id, or email, and update profile fields.
“Add john@example.com to Enterprise Prospects.”Two habits worth knowing: it declines to quote a rate on a sample too small to mean anything rather than quoting it anyway, and if the documentation does not answer a question it offers to open a support ticket instead of guessing.
How it works
Connect it in three steps
In Personyze, open Integrations → AI Assistants (MCP) and copy the server URL.
https://personyze.com/mcp/CopyOn Claude it installs from the connector directory in one click. On ChatGPT and other clients, paste the URL as a custom connector.
A Personyze screen opens and you tick what it may reach. The assistant gets its own credential — nothing to copy, and revocable whenever you like.
The tools
Everything the panel does, as tools it can call.
Campaigns and the content inside them, analytics, audiences and visitor profiles, goals and placeholders, docs search and support tickets — each one a tool the assistant can call directly, with your account’s permissions and nothing copied between tabs.
Build it and change it
Open a draft, describe the audience in plain words, write or rewrite the action, and push it to staging. Nothing goes live until somebody says so.
Read what happened
Per-campaign results over any date range, where traffic is leaking, how the catalogue is performing — and a report on a schedule so nobody has to ask again.
Manage who sees it
Build and edit audiences, look up a visitor and write fields back, define goals, placeholders and page groups, and check the tag is actually reporting.
These are the ones it uses most. Anything they do not cover falls through to personyze_api — the same REST objects any API client uses — so the assistant is never boxed in by the tool list.
Safety
Guardrails that are structural
Giving an assistant write access to your website deserves scrutiny. These are properties of how it works, not a warning label.
01
New work is staged
Campaigns an assistant creates are shown only to visitors arriving from a staging IP. Real traffic keeps seeing what is live, and you publish in the panel. Targeting described in plain words is compiled into real rules and read back to you.
02
Live changes are confirmed
Pausing or resuming an action names the action, the campaign it sits in, and what visitors stop or start seeing — one action at a time, and fully reversible.
03
Access is revocable
Delete the assistant’s key under Integrations to disconnect. Revocation is immediate, and the connector only ever acts with the permissions of the account behind it.
04
Standards-based auth
OAuth 2.1 with authorization code and PKCE, dynamic client registration (RFC 7591), and discovery per RFC 9728 and RFC 8414. Scope personyze:full.
Ask your assistant what to fix this week.
Connect once and your personalization data, campaigns and audiences are reachable from the assistant you already use — with staging and confirmation guardrails built in.