pinksheep
Guides/Implementation

MCP Server for Salesforce: connect Salesforce to AI agents with approval before updates

Quick answer

A Salesforce MCP server lets AI tools query your CRM, read opportunities, and propose updates without writing the MCP layer yourself. Choose what agents can do, keep approval before important writes, and stay in control of what changes in Salesforce.

A Salesforce MCP server lets AI tools query your CRM, read opportunities, and propose updates without writing the MCP layer yourself. Choose what agents can do, keep approval before important writes, and stay in control of what changes in Salesforce.

9 min readUpdated 24 March 2026

What an MCP server for Salesforce does

An MCP server for Salesforce gives AI tools a clean way to use your CRM through MCP. Instead of building a custom integration for every AI client, you expose the Salesforce actions your team actually wants agents to use.

That means agents can query records, review pipeline data, and prepare updates inside the limits you set, without opening up your whole CRM.

Common operations exposed by the Salesforce MCP server:

  • CRM reads. Query leads, contacts, accounts, opportunities, and the context your team needs to work with.
  • Selected updates. Propose field changes, owner changes, task creation, or stage updates, then review them before they are written.
  • Pipeline visibility. Pull the records, owners, and stage data needed for reporting, cleanup, or follow-up work.
  • Cross-tool use. Use the same Salesforce MCP server inside AI tools like Cursor or Claude instead of rebuilding the connection each time.
  • Full visibility. Keep a record of what the agent read, what it proposed, and what your team approved.

Why MCP for Salesforce instead of direct API access

The Salesforce API is powerful, but an MCP server is often the easier path when you want AI tools working inside CRM data without building a custom integration for every client.

ConcernDirect APIMCP Server
Scoped accessYou manage direct API access yourselfYou choose the objects and actions agents can use
Approval before writesNo review layer by defaultYou can require review before important CRM changes
Audit trailYou add your own loggingAgent activity stays visible in one place
Reuse across AI toolsCustom work per client or integrationSame server can be reused across MCP-compatible tools
Time to working setupMore engineering work up frontFaster path for common CRM access patterns

How to set up the Salesforce MCP server

The setup is simple: connect Salesforce, decide what actions agents can take, choose what needs review, and publish the server to the AI tools you already use.

1

Connect Salesforce

Connect your Salesforce org and choose the objects and fields the server should be able to see.

2

Choose the actions

Pick the reads, searches, and selected CRM updates you actually want agents to use.

3

Set approvals

Decide which field changes, owner updates, task creation, and other writes should be reviewed before they run.

4

Publish to your AI tools

Copy the MCP server URL into the AI tools you want to use, like Cursor or Claude.

5

Review activity

Check what agents read, what they proposed, and what was approved so you can keep access tight over time.

Common Salesforce jobs agents can help with

The table below shows common CRM jobs teams often hand to AI with review built in before important record changes.

Agent use caseWhat the agent doesReview
Pipeline hygieneReviews stale opportunities and prepares the next update or follow-up actionReview before record changes
Lead routingFinds the right owner or queue and prepares the assignment for approvalReview before owner changes
Activity loggingPulls call or deal context and drafts the task or note your team wants to saveReview before writes
Forecast reportingReads the pipeline and drafts a report your team can review before sharingUsually read-only
CRM cleanupFinds missing or inconsistent fields and prepares proposed updatesReview before updates

When to use MCP vs direct Salesforce API

Use the MCP server when you want AI tools working inside Salesforce with approvals and clear limits. Use the direct API when you are building a single-purpose integration and want to own the implementation end to end.

  • Use MCP when: You want a faster path to CRM access for AI tools, you need review before important writes, and you want the same server reused across multiple MCP-compatible clients.
  • Use direct API when: You are building a narrow integration, you want full implementation control, and you do not need an approval layer for AI actions.

Frequently asked questions

What is the difference between an MCP server and a Salesforce API wrapper?

A Salesforce API wrapper is a direct integration. A Salesforce MCP server gives AI tools a controlled way to query CRM data and propose updates inside the limits you set.

Do I need to build the MCP server myself?

No. Connect your Salesforce org in Pinksheep, choose what objects and fields agents can use, and publish the MCP server URL to the AI tools you want to work in.

Can the MCP server work with Einstein AI?

Yes. Einstein works inside Salesforce, while an MCP server lets other AI tools use Salesforce data with the limits and approvals you choose.

How do approval gates work with MCP?

You choose which writes need review. Your team sees the proposed change and the context before it runs, then approves or rejects it.