MCP for Salesforce: Connect Your CRM

TL;DR

MCP Salesforce integration lets AI read and write your CRM data in real time.

Your AI assistant can pull contact records, update opportunities, and log activities without copy-pasting.

Setup involves an MCP server, a Salesforce connected app, and proper OAuth scopes.

Security is manageable. You control what the AI can see and do.

This is practical today. You do not need to wait for Salesforce to ship a native solution.

What is MCP and why does it matter for Salesforce?

MCP Salesforce integration is the answer to a problem most sales teams have right now. Your AI tools are locked out of your CRM.

MCP stands for Model Context Protocol. It is an open standard that lets AI models talk to external tools and data sources. Think of it as a shared language. The AI speaks it. Your Salesforce org speaks it. They can finally have a proper conversation.

Without MCP, your AI assistant knows nothing about your pipeline. You paste in contact details. You copy out meeting notes. You summarise opportunity stages by hand. It is slow and it breaks the flow. With MCP, your AI can read a contact record before a call, update a deal stage after a meeting, and log a note straight into Salesforce. All from inside your AI chat interface.

If you are new to MCP, start with the Model Context Protocol guide. It covers the basics before we get into Salesforce specifics.

The AI does not replace your CRM. It just stops ignoring it.

What can you actually do with MCP and Salesforce?

Can AI read live Salesforce data?

Yes. That is the starting point. An MCP server sits between your AI model and Salesforce. When the AI needs data, it calls the MCP server. The server talks to Salesforce via the REST API. The AI gets the answer.

Here is what that looks like in practice. You ask your AI: "What is the status of the Acme deal?" The AI calls the MCP server. The server queries Salesforce. It returns the opportunity stage, close date, and last activity. The AI gives you a plain-English answer in seconds. No tab switching. No manual lookups.

Can AI update Salesforce records?

Yes, with the right permissions. You can give your AI tools write access to specific objects. Common use cases include logging a call note after a meeting, moving an opportunity to the next stage, creating a new contact from an email, and updating a lead status after a follow-up. You control the scope. The AI only does what you allow it to do.

What about reports and forecasting?

This is where it gets useful for founders and sales managers. You can ask your AI to pull a forecast summary, identify stale deals, or flag contacts who have not been touched in 30 days. Instead of building a report in Salesforce and exporting it, you just ask. The AI queries Salesforce, crunches the data, and gives you an answer. You can see similar patterns in how MCP connects to other business tools. The MCP for Power BI post covers the reporting angle in more depth.

The CRM becomes a tool the AI can use, not a database you have to query yourself.

How do you set up MCP for Salesforce?

Step 1: Set up a Salesforce connected app

You need a connected app in Salesforce to give external tools API access. This is standard Salesforce OAuth. Go to Setup, search for App Manager, and create a new connected app. Enable OAuth. Select the scopes you need. At minimum you will want access to your data via APIs and the ability to perform requests at any time. Save the consumer key and secret. You will use these in the next step.

Step 2: Deploy an MCP server

This is the middleware layer. The MCP server translates between your AI model and the Salesforce API. You have a few options. You can build a lightweight server using the MCP SDK. You can use a pre-built Salesforce MCP connector. Or you can work with a team like Devwiz, who build custom CRM and MCP integrations and can wire this up for you. The server needs to accept tool calls from the AI model, authenticate with Salesforce using your OAuth credentials, map tool calls to Salesforce API endpoints, and return structured data the AI can use.

Step 3: Define your tools

The MCP server exposes tools to the AI. A tool is a named function the AI can call. You define what each tool does. For Salesforce, your tools might include: get a contact by name or email, get opportunity details by name or account, update an opportunity stage, log an activity on a record, and search accounts matching a term. Keep tools focused. One tool does one thing. This makes it easier for the AI to use them correctly.

Step 4: Connect to your AI model

Most enterprise-grade AI models support MCP. Claude, GPT-4, and others can connect to MCP servers. You point the model at your server. It discovers the available tools. It starts using them. You can see the same pattern applied to internal tools in the Claude MCP internal tools guide.

A focused tool library beats a swiss army knife every time.

Is MCP for Salesforce secure?

How do you control what the AI can access?

This is the right question to ask before you start. MCP does not give AI blanket access to Salesforce. You define the tools. You set the OAuth scopes. You control the blast radius. Start narrow. Give the AI read access to the objects it needs. Add write access to specific fields only. Review the tool definitions before you go live. You can also use Salesforce profiles and permission sets to restrict what the API user can see. The AI inherits those restrictions.

What about sensitive data?

Salesforce holds sensitive data. Contacts, revenue figures, pipeline values. You do not want that leaking into an AI model that logs everything. Two things help here. First, use a self-hosted or private MCP server. Your data does not leave your infrastructure. Second, apply field-level security in Salesforce to exclude fields the AI does not need. If you are in a regulated industry, talk to your security team before you start.

What about audit trails?

Salesforce logs API calls. You can see every query and write the MCP server makes. That gives you a full audit trail. If something goes wrong, you can trace it. This is actually an advantage over manual data entry. AI actions are logged. Human copy-paste errors are not.

How does this fit into a broader AI strategy?

MCP Salesforce is one piece of a bigger picture. The real value comes when your AI can move across systems. CRM to calendar. CRM to email. CRM to project management. That is what AI agents do. They work across tools, not just inside one. The AI agents for business guide covers how multi-tool agents work in practice. MCP is the plumbing. Once it is in place, adding more integrations gets easier.

FAQ

What Salesforce editions support MCP integration?
Any edition with API access supports it. That covers Professional, Enterprise, Unlimited, and Developer editions. Essentials edition does not include API access by default.

Do I need a developer to set this up?
For a production setup, yes. You need someone comfortable with OAuth, REST APIs, and either Python or TypeScript. Teams like Devwiz can handle the full build if you do not have in-house capability.

Can MCP write to Salesforce, or is it read-only?
Both are possible. Read-only is simpler and lower risk. Write access requires more careful scoping but opens up the most useful automation use cases like logging calls and updating deal stages.

How long does a typical MCP Salesforce integration take to build?
A basic read-only integration with five to ten tools takes a few days. A full read-write integration with custom tools and security review takes one to two weeks depending on complexity.

Ready to connect your CRM?

MCP Salesforce integration is not a future feature. You can build it now. The protocol is open. The tools exist. The main thing stopping most teams is knowing where to start. If you want to see what this looks like for your specific setup, book a free AI transformation call. We will map out what is worth connecting and what to leave alone.