Connect Salesforce to AI agents over MCP (with Stepper Skill Sets)

Let your AI agent find, create and update Salesforce records through a Stepper Skill Set, limited to the actions you choose.

A Skill Set is a toolkit you build in Stepper: you connect your apps, pick which actions an agent may call, and Stepper serves them as one MCP endpoint. The agent never sees your Salesforce password or API keys, and every call is logged.

Last checked:

What you can let an agent do in Salesforce

These are the Salesforce actions you can add to a Skill Set. They're the same actions Stepper workflows use, listed on Stepper's Salesforce page. You choose which ones the agent gets.

For example, you could ask

  • Find the lead for [email protected] and add her to the Spring Workshop campaign.
  • Create a contact for the person who emailed about wholesale, then update the account's phone number.
  • Create Record Creates a new record of a specified Salesforce object.
  • Update Record Updates an existing record for a specified Salesforce object.
  • Delete Record Deletes an existing record of a specified Salesforce object.
  • Create Contact Creates a new Contact in Salesforce.
  • Update Contact Updates an existing Contact in Salesforce.
  • Create Lead Creates a new Lead in Salesforce.
  • Update Lead Updates an existing Lead in Salesforce.
  • Add Contact to Campaign Adds an existing Contact to an existing Campaign.
  • Add Lead to Campaign Adds an existing Lead to an existing Campaign.
  • Send Email Sends an email using Salesforce Simple Email Actions.
  • Find Record Finds a record of a specified Salesforce object by a field and value you choose.
  • Find Record by Query Finds a record of a Salesforce object using a SOQL WHERE clause.
  • Make HTTP Request Make an HTTP request to any URL with full control over method, headers, and body.

Set it up in 5 steps

You need a Stepper account, a Salesforce account, and an AI client that supports remote MCP servers, such as ChatGPT, Claude, Cursor or VS Code. Your own agent works too, using a Skill Set token.

  1. Connect Salesforce to Stepper

    In Stepper, connect your Salesforce account. Use an account that can do what you want the agent to do, and no more.

  2. Create a Skill Set

    Go to Skill Sets in Stepper, create a Skill Set for this agent, and use Add Entry to add your Salesforce connection.

  3. Restrict the actions

    Choose only the Salesforce actions the agent needs. An entry with no actions chosen allows every Salesforce action, so a short list is safer. To pin a value, such as one channel, table or calendar, hardcode that parameter so the agent can't change it.

  4. Add the Skill Sets server to your AI client

    Add this URL as a remote MCP server:

    https://mcp.stepper.io/skill-sets/mcp

    Then sign in to Stepper when your client asks. On the Stepper screen that follows, pick the Skill Set this agent should use. To switch to a different Skill Set later, remove the server and connect again.

    Chat apps (ChatGPT, Claude and others): add it as a custom connector or app, wherever your client lists remote MCP servers.

    Code editors (Cursor, Windsurf and others that use an mcpServers config):

    {
      "mcpServers": {
        "stepper": {
          "url": "https://mcp.stepper.io/skill-sets/mcp"
        }
      }
    }

    Your own agent or script: generate a Skill Set token in Stepper and send it as Authorization: Bearer sst_….

  5. Test it

    Ask the agent what it can do (it checks the Skill Set's list of skills) and make sure only the Salesforce actions you chose appear. Then run one request using the lowest-risk action you allowed, such as a lookup or a draft, and check the call in the Skill Set's call log in Stepper.

Salesforce's own MCP server, and when to use it instead

Salesforce runs its own hosted MCP servers. They connect an AI client to your org with OAuth, and a Salesforce admin turns them on in Setup. Read Salesforce Hosted MCP Servers for setup.

Use Salesforce's servers when the agent works mostly in Salesforce, needs objects beyond the actions listed above, and should follow your org's own permission model. You'll need an admin to set them up.

A Skill Set fits when the agent needs Salesforce alongside other apps through one endpoint, or when you want to choose exactly which actions it can call, lock parameter values, and check a log of every call.

Guardrails

An agent can misread a request, and some Salesforce actions change or remove real data. Set these up before you rely on it.

Scope

Leave Delete Record and Find Record by Query off unless the agent needs them. If it only works with leads, add the lead actions and skip the general record actions.

Use a separate Skill Set for each agent or role, so each one gets only its own actions and tokens.

Audit

Every skill call is logged with the tool name, status, how it authenticated, and when. Call logs are kept for 30 days and are visible in Stepper, so you can check what the agent did in Salesforce.

Revoke

Revoke a Skill Set token from Stepper to cut off a script or client straight away. To stop an agent using one action, remove it from the Skill Set; that applies from the agent's next call, with no need to reconnect. You can also remove the server from your AI client.

Add with care: Delete Record, Make HTTP Request. Make HTTP Request can call any URL with your connection, so it gives the agent far more reach than the other actions.

Connect other apps to AI agents over MCP:

Salesforce is a trademark of its owner. Stepper isn't affiliated with or endorsed by Salesforce; its name is used here only to describe the integration.