MCP: what to hand over to an AI assistant in OneTrace.pro
How to connect Claude, ChatGPT, Claude Code or Cursor to a project over MCP, what the assistant can do with your data, and 15 scenarios for marketing, analytics, support and development.
AI assistants write good copy, but until recently they knew nothing about your customers: to ask about campaign revenue, you had to export a report and paste it into the chat. MCP (Model Context Protocol) is an open standard that lets an assistant reach external systems on its own. OneTrace.pro has an MCP server, so Claude, ChatGPT, Claude Code and Cursor work with project data directly — and not only read it, but act on it.
How to connect
The server address is the domain you sign in to the platform on, plus the /mcp path.
- Claude.ai and ChatGPT connect through your account: add a custom connector with the server address, sign in as usual and pick a project. No keys needed.
- Claude Code, Cursor and other developer assistants connect with a secret project key with the permissions you choose — ideally a separate key for each assistant.
One connection works with one project. The step-by-step guide is in the help center, in the “AI assistants (MCP)” article.
What the assistant can use
The MCP server’s tools are the operations of the public API: reports and funnels, profiles and events, segments, templates, campaigns, journeys, loyalty, promo codes and transactional messages. Responses and limits are the same as in the API, so the assistant cannot do anything the API cannot do with the same permissions.
Besides the operations, the assistant has a reference: the formats of segment rules, journeys, reports and templates, plus the events, traits, integrations and templates of your specific project. So when you ask it to “build a segment of people who abandoned a cart”, it does not make up an event name — it takes it from the project catalog.
Scenarios
Analytics without an analyst
- Revenue questions. “How much revenue did campaigns bring in the last 30 days? Compare with last month.” The assistant runs an attribution report and explains the difference.
- Finding the bottleneck. “Show the funnel from product view to order by traffic source. Where do we lose the most?”
- Retention. “What share of September buyers came back for a second purchase? Compare with August.”
- A dashboard for management. “Build a dashboard: weekly revenue, new and returning buyers, the contribution of campaigns and recommendations.” It can then be emailed on a schedule.
Segments and campaigns
- A segment from a description. “Build a segment of customers who ordered more than twice but haven’t visited the site for two weeks.” The assistant writes the rules and shows the audience size.
- Preparing a campaign. “Draft a campaign for this segment: an email about the new collection, an A/B test of two subject lines, sent on Thursday at 10:00 in the recipient’s time.” It launches only after your confirmation.
- Reviewing a campaign. “Why won’t the ‘Autumn sale’ campaign start?” or “Which variant won in the last campaign and by how much?”
Journeys
- A journey from a description. “Create a journey: a welcome email after sign-up, check for a purchase two days later, and if there is none, send an email with a code from the welcome pool.” The assistant creates a draft, validates it and publishes only with your approval.
- Auditing scenarios. “Go through the active journeys: where is goal conversion below 2%, and at which step do most participants drop off?”
Support and customer care
- A customer card in the chat. “What happened with
[email protected]over the last week — which emails did she get, what did she buy, how many points does she have?” Without the personal data permission, email and phone are masked. - Message status. “Did the confirmation code sent 10 minutes ago reach the customer, and through which channel?”
- A goodwill gesture. “Add 500 points to this customer with the reason ‘delivery delay’” or “Issue them a code from the apology pool.” The assistant does either only after confirmation.
Development and rollout
- Checking the integration. In Claude Code or Cursor: “Check that checkout_started events arrive with the cart_value and items properties, and show the latest events of the test profile.”
- The event catalog. “Which events and properties does the project already have? What is missing for a browse abandonment journey?”
- Migration. “Here is a description of our scenarios from the old platform — build drafts of the same journeys.” Together with AI email import, this shortens a migration from weeks to days.
Security
An assistant’s access to customer data is a matter of trust, not just convenience, so the limits are built into the server rather than left to the model’s judgment.
- Your role’s permissions. When signed in through your account, the assistant gets your role’s permissions in the project. Change the role and the permissions change at once; remove the user from the project and the connection stops working. With a key, the key’s permissions apply.
- Confirmation before anything irreversible. Launching a campaign, publishing a journey, deleting, adjusting points, issuing promo codes and sending events or messages are marked as actions the assistant asks a human about first.
- Personal data masking. Email, phone and other personal data are visible in clear text only with the “Personal data without masking” permission.
- Audit log. Every change shows up in the project log marked “via MCP”, as do the connection itself and its revocation.
- One-click revocation. Assistant connections are listed in your profile — with the project and the last use time — and revoked with the Revoke button.
Which plan
The MCP server is available on the Growth and Enterprise plans and during the trial. On Start the assistant connects and sees the list of capabilities, but every action returns a message that Growth is required. More on the AI assistants page and in pricing.
The easiest way to try it is on your own data: 14 days of Growth for free, and connecting the connector takes a minute. If you run marketing with a small team, also see the Marketing with a small team and AI solution.