Ontraport's MCP server connects an AI tool to your account, so you can ask about records and carry out supported tasks. Let's connect the right account, control access, and try a lookup and a reviewed change.
MCP is a standard way for AI tools to work with software. Choose a client that supports remote MCP servers over streamable HTTP. In that client's connection settings, add mcp dot ontraport dot com and sign in. The Support Center has client-specific instructions.
Before authorizing, check the account number and email on the authentication screen. If either is wrong, choose Use a different account. This matters when you manage several accounts. Authorize only when both match the account you intend to use.
Sign-in uses OAuth, the recommended connection method. API keys are another option for custom tools and automation platforms. Key-based access stays within that key's permissions. Treat the key like a password; don't paste it into a conversation.
An OAuth connection inherits your Ontraport role permissions. Separately, your AI client can offer switches for individual tools. Turn off tools you don't need, especially destructive or payment actions. This connection can change or delete records and charge cards. Switching off a tool in one client isn't an account-wide block on other connections.
To stop using an OAuth connection, remove or disconnect it in your AI client's connection settings; names vary by client. For API-key connections, disable or delete the related API credentials in Ontraport. Removing one connection doesn't remove other connections you set up.
Before you start, add useful descriptions where available. Explain what an object stores, what belongs in a field, and what a relationship means. For example, describe whether a date means the first inquiry or the most recent purchase. Those are different questions, even when both fields contain dates. These descriptions help teammates understand your setup. Ontraport explicitly says an optional, internal Site description also helps AI understand the site's purpose. That doesn't establish that AI reads every other description.
Start with a small lookup, like up to twenty contacts added this week. Returned record details are paginated and the AI can only hold a limited set at once. A page of contacts isn't your whole database.
Counting is different. The server's count objects tool counts all matching records without pagination, and can return sums or averages for fields configured for totals. Check the filters, permissions and field meanings before trusting the answer. Don't count a sample of returned records and call it an account total.
For inspecting every record or making bulk changes, use an API or workflow with explicit pagination and checks for missing or repeated records. A server count doesn't mean the AI inspected every record. Track the record IDs from each page, use a consistent ordering, and reconcile the unique records inspected against the intended selection. If records change during the run, recheck the selection before writing. Stop when a page fails rather than silently continuing with an incomplete set. Validate each proposed write.
For your first change, ask for one contact and one tag, with a review before writing. Confirm the record and tag, then check the result in Ontraport. The example email is a placeholder. Connect carefully, distinguish counts from record inspection, and review changes. Next up: where your data lives in Ontraport.