You can now connect Claude, ChatGPT or any other MCP-capable assistant straight to your Personyze account and work in plain conversation — ask how campaigns are doing, find where you’re losing traffic, build a campaign, pause an action, manage audiences. No exports, no copying between tabs, no code.
One connection, and your assistant can read and act on your Personyze account.
What MCP is, briefly
MCP (Model Context Protocol) is the open standard AI assistants use to work with outside services. Anthropic introduced it in November 2024; OpenAI adopted it in 2025, and by early 2026 there were over 10,000 active public MCP servers. The usual analogy is “USB-C for AI”: instead of every tool building a bespoke integration for every assistant, a tool exposes one MCP server and every compatible assistant can use it.
Personyze runs one. An assistant that connects to it reaches your account directly, under your account’s permissions — the same external API surface as the full-featured API.
Server URL: https://personyze.com/mcp/
Connecting an assistant
Most assistants ask only for the URL and then walk you through signing in:
- In Personyze, open Integrations → AI Assistants (MCP) and copy the Server URL.
- In the assistant, add a custom connector and paste it — in Claude: Settings → Connectors → Add custom connector; in ChatGPT: install it from the ChatGPT app directory instead, which skips the pasting.
- Approve the connection on the Personyze screen that opens. Nothing to copy and paste — the assistant receives its own credential.
The AI Assistants (MCP) card in Personyze — copy the Server URL into your assistant. API keys are only for tools that can’t open a browser.
Tools that can’t open a browser — a script, an IDE plugin — can send a Personyze API key as a header instead:
Authorization: Bearer YOUR_API_KEY
A note on availability: Personyze is now listed in the Claude connector directory, so on Claude it is a one-click install rather than pasting a URL — see the announcement. On ChatGPT it installs from the ChatGPT app directory the same way; any other assistant connects with the URL above as a custom connector. Both Claude and ChatGPT gate connectors by plan tier, so check yours if the connection screen never appears.
What you can ask it to do
The wording is yours — these are kinds of question, not fixed commands.
Understand how campaigns are performing
- “Which campaigns performed best in the last 30 days?”
- “How is the pricing-page banner doing compared to April?”
- “Which campaigns had no activity at all last month?”
It reads the daily performance summary per action — impressions, deliveries, clicks, dismissals, failures, your configured goals and revenue totals. It will decline to quote a click-through rate on a sample too small to mean anything and say so, rather than quoting it anyway.
Find where you’re losing traffic
- “Which landing pages lose the most sessions?”
- “Which traffic sources convert worst?”
- “What organic search terms bring people in?”
Site-wide traffic for any date range by page, landing page, referrer, search term, device, city or country — with sessions, bounces, goals, goal value and time on page, plus the personalization impressions and clicks on those pages. Rates are computed after grouping and low-traffic rows are excluded when ranking by a rate, so a 100% bounce rate over three sessions doesn’t outrank a real problem.
See what’s running, and why
- “Why isn’t the welcome banner showing?”
- “What is the targeting on campaign 1834?”
- “Which of my actions are live and which are still in testing?”
Targeting comes back as readable rule text — “Include on current page: Contains /pricing and not Full URL contains login” — not raw configuration.
Build and change campaigns
- “Create a campaign targeting returning mobile visitors.”
- “Rename it to Summer Promo and re-target to visitors from paid search.”
- “Pause the popup on the checkout page.”
Work with audiences and profiles
- “How many audiences do I have, and how big is each one?”
- “Add john@example.com to the Enterprise Prospects audience.”
- “Which visitors have visited more than five times?”
Ask how the product works — or get a human
“How do I set up conversion tracking?” is answered from the knowledge base with links to the articles used; if the docs don’t cover it, the assistant says so and offers to open a support ticket rather than guessing — writing up what you were doing and what it already checked, so support doesn’t start from nothing.
The tools behind it
23 tools, plus a fall-through to the REST API for everything else.
Twenty-three tools cover campaigns and actions, analytics, audiences and visitor profiles, placeholders, docs search and support tickets. Anything the dedicated tools don’t cover falls through to personyze_api — the same REST objects available to any API client — so the assistant isn’t boxed in by the tool list.
Safety: staging, confirmation, revocation
Giving an assistant write access to your website deserves scrutiny. Three properties matter here:
- New campaigns are created in staging mode. They’re shown only to visitors arriving from a staging IP; real traffic keeps seeing what’s live, and you publish in the panel. Targeting described in plain words is compiled into real rules and read back to you — because a rule that says something slightly different from what you asked looks identical from the outside.
- Pausing or resuming an action is live, so it’s confirmed first — naming the action, the campaign it sits in, and what visitors stop or start seeing. One action at a time, and fully reversible.
- Access is revocable instantly. Every connected assistant rides on an API key belonging to your account and acts with that account’s permissions. Delete the key under Integrations to disconnect; revocation is immediate.
For developers building a client rather than using a ready-made assistant, the server implements OAuth 2.1 (authorization code with PKCE/S256, public clients), dynamic client registration (RFC 7591), and discovery per RFC 9728 and RFC 8414, with scope personyze:full.
Why this matters for personalization
Personalization tools have always had a reporting problem: the data lives in one panel, the decision happens in someone’s head, and the change happens in a third place a week later. MCP collapses that. You can ask which pages leak sessions, get the answer from your real personalization data, and stage a campaign against that page — in the same conversation, without leaving your assistant.
It also means Personyze fits into whatever agentic stack you’re already assembling. Forrester expects 30% of enterprise app vendors to launch MCP servers in 2026; the practical upshot is that your assistant can read your CRM, your ads and your personalization platform side by side, and reason across them.
Docs
- MCP server documentation: full tool list, auth details and troubleshooting.
- Server URL:
https://personyze.com/mcp/
Originally published on the Personyze blog.
Top comments (0)