With Salesforce Catalog, agents stop guessing how your org works

The average customer satisfaction (CSAT) score your service team tracks fell this week. Before anyone decides whether customers are less happy, someone has to answer a plain question; where does the number come from? Usually that means a ticket to whoever built the data pipeline, and a wait. The answer is already in your org, in the objects, fields, flows, and calculated insights that record how your business runs. AI assistants and agents know Salesforce. It doesn’t know your org. So, it guesses. Salesforce Catalog gives people and agents a way to look up how your org is built before they answer, build or change anything. Most questions about how the org works come from people who don’t write code. It’s your service lead checking a number, the analyst choosing between two definitions, the data owner about to change a field, the admin cleaning up after someone leaves. Getting an answer today means digging through Setup, reading a data model diagram or finding the person who built it. An agent that is not grounded in your org’s working won’t hesitate. It answers and sounds certain, even when it’s wrong. What Salesforce Catalog is Salesforce Catalog is an up-to-date index of how your org is built. It updates as your team deploys changes, so you don’t have to write or maintain documentation for it. It covers Objects, fields, flows, Apex, permission sets, users, and more. It also covers how your Data 360 org works, from data streams to calculated insights, segments, and more. To use Catalog, you ask an AI assistant or agent a question, and it looks in the Catalog before it answers. It also understands how objects and your processes are connected and is permission-aware of your Org’s policies. Every question runs as the signed-in user. Trace where a number comes from, and what a change would break Maya is a service operations manager at a home appliance company. She notices a drop in the average CSAT score on her team’s dashboard. She asks three questions in the AI assistant she already uses, and it calls Salesforce Catalog before it answers. “Where does the average CSAT measure get its data?” The assistant traces it step by step, from where it was uploaded to the calculated insight that averages it per customer. “Can we get rid of the Severity field? Hardly anyone fills it in.” The Open High Severity tickets measure in a live insight depends on it, so the agent flagged it for sign-off instead of suggesting the change. “How is the customer lifetime value calculated?” She wants to know which customers to call back first. The assistant found two definitions in one insight, one loaded once onto the customer record and one computed from actual orders, and shows where each came from. She goes with the one computed from orders. Without Catalog, each of these would have meant finding help, an engineer who built the data pipeline, the analyst who owns the insight, or an afternoon clicking through Setup. Maya was able to get all her answers quickly, and from the assistant she already uses. Each answer was limited to what she was entitled to see. Figure 1 – Illustrative example. Answers to your questions grounded in your metadata Questions like these come from every part of the business. How it finds what you asked about The Catalog finds things by meaning as well as by name. It compares the question with the names, labels, and descriptions of everything in your Org, and it matches on the meaning as well as the exact words. So, “whether an order arrived” finds Shipment.Delivery_Status__c, even though the word “arrived” isn’t in the field. It sends back a short list of the closest matches, best first, and the agent reads the details of only those. That matters in a large org with thousands of objects and fields. The agent responds quickly since it is based on the few items that actually fit. Fewer guesses, and fewer failed deploys for Developers You ask your coding agent for a quick change to a renewal flow. The code compiles and reads well, but it points at Contract_End_Date__c, a field your org has but no longer uses. The real one is Term_End__c, added a few months ago by a team that’s since moved on. You find out when the deploy fails, and the next half hour goes to working out which of the two contract fields is the one still in use. With Catalog, the agent looks up the real names first. Developers write their requirements in plain language and get a response grounded in their org’s metadata, limited to what the signed-in user is allowed to see. It also saves the agent from exploring a large org within the Salesforce CLI, or from working off a pasted metadata export. The CLI reads thousands of components the agent doesn’t need, and an export is out of date after the next deployment. Catalog does the searching on its side and hands back only the relevant items that match. Getting Started Salesforce Catalog is available in every edition that supports Data 360. An admin turns it on in Data Cloud Setup and turns on the MCP server. You can then add it to your AI assistant and use and sign in with your Salesforce account. The setup guide under Resources has the steps and permissions. Agentforce Vibes comes set up. Claude, ChatGPT, Cursor, or other AI assistants connect and sign in with a Salesforce account. They all read the same Catalog as the signed-in user, so a team that uses more than one AI tool gets the same picture of the org. Teams building their own agents can use the Data 360 MCP server. Agents that need shared context and memory can use Agent Context Engine, which uses Catalog to find the right context for each job. Resources Salesforce Catalog (Salesforce Help) Metadata Grounding MCP Server (Beta) (Salesforce Developers) Turn on Salesforce Catalog Any unreleased services or features referenced in this or other press releases or public statements are not currently available and may not be delivered on time or at all. Customers who purchase Salesforce applications should make their purchase decisions based upon features that are currently available.