Salesforce MCP: what is it, what can it do and is it safe?

An MCP (Model Context Protocol) is a kind of universal power socket that lets an AI connect to another application, such as your email, your calendar or, in this case, Salesforce.

You work from your AI with your real, current data, instead of having to copy, paste or explain everything yourself. It looks things up on its own, and it can read, write and delete in Salesforce without you logging in to the main screen yourself.

Retrieving and logging information

Ask: “Who is the contact person at customer Y, and what is our relationship history?” You immediately get the name, job title and contact details back, followed by the latest conversations, active deals and previous cases. Useful when you’re new, and just as useful when you simply don’t have that customer top of mind.

After a customer conversation, you say: “Log this conversation on account X, next step is to send a quote.” Your CRM is updated without you having to log in, and admin stops being a separate task you keep putting off. You can also just dictate it, for example on your phone.

Build dashboards in no time

Preparing the quarterly figures for the management team? Ask: “Make a chart of the revenue trend by region over the last four quarters.” You get a chart back right away that you can drop straight into your presentation. That means you don’t have to wait until the BI department has time for you.

To answer that, we look at three points.

What data can your AI see? An MCP doesn’t give your AI any new permissions: you connect it to a Salesforce user, with the permissions that user already has. Whatever that user can see and do, the AI can too.

What can the AI do with it? You decide which tools to switch on: you can limit your AI to searching and reading, or also let it write and delete. Our tip: start with read-only.

What happens to your data at your AI vendor? The data you request goes to your AI vendor. That can be technically secure, but that doesn’t tell you whether it’s permitted within your organisation. What the AI vendor is then allowed to do with your data differs per subscription: some use your messages to train the model, others don’t. Find that out before you roll this out widely.

We help clients with these trade-offs every day and are happy to think it through with you.

Setting up a Salesforce MCP is fairly simple. Read the steps here, or watch the video at the top of this page (the setup starts at 5:00).

What you see above is a single connection: Salesforce to your chat. Your chat then only has access to your Salesforce data. Want to query multiple sources at once, such as your ERP or your ticketing system? Then read our blog post on MCP architecture.

Want to get started with this yourself, or want to know what it means for your organisation? Go to our contact page and leave your details, and we’ll get in touch with you.