Connecting an AI client to a Salesforce org is now a short setup task. An admin activates the MCP servers, creates an External Client App, and a user authorizes the connection from Claude. From that moment the agent operates with exactly the permissions of the person who authorized it. Every tool call respects that user’s object permissions, field level security, and sharing rules, and every action carries that user’s name in the audit trail.
That design is sound, and it is also the problem. Most orgs have never audited what one user can reach. The permission set granted for a 2022 project is still assigned, and the profile cloned from System Administrator is still in use. None of that mattered much when access meant a person clicking through screens. It matters when an agent queries the same surface in seconds.
This session treats the MCP connection as an access review trigger rather than an integration task. It examines the controls Salesforce ships for this purpose, including read only servers, Named Queries, and restrictions on which profiles can connect.
Attendees will leave with an access review sequence to run before the first connection is authorized, and the API log filter that shows what agents did afterward.
This is an interactive session. Bring your burning questions about Salesforce security and get them answered live by guest speaker Mike Rogan, Founder and CTO of Wayfinder, and host Matt Meyers, CTA, CoFounder and CEO of EzProtect.
Register for this event and we will see you online.