Salesforce MCP Server Setup Guide for Claude, ChatGPT, and Cursor

This guide covers the general process of setting up a Salesforce MCP server. It walks through connecting your Salesforce org, configuring your AI tool of choice, and verifying the connection is live. A Salesforce MCP server is the bridge that creates an MCP Salesforce integration between an AI tool and your org — once the connection is active, Claude or ChatGPT can read your org structure and execute configuration tasks through natural language conversation.

These steps work with any Salesforce MCP server. Cirra AI is used for the screenshots and examples below, and covered in more detail at the end. The process takes under 10 minutes for a first connection.

What You Need Before Setting Up a Salesforce MCP Server

Choosing an AI Client: Claude vs ChatGPT vs Cursor

All three connect to a Salesforce MCP server the same way at the Salesforce end. What differs is the setup method inside each client, and the plan you need.

If you're not sure which to pick: use whichever your team already has a paid seat for. Claude is the lowest-friction starting point if nobody has a preference, since it works on a free plan and the setup is entirely UI-driven. Cursor makes the most sense if the person doing Salesforce admin work is also writing code day to day.

Gemini Enterprise also supports MCP, through a custom connector rather than a built-in one. It's a more involved setup than the three above, so it isn't covered step by step in this guide.

The account and Salesforce org connection below is the same no matter which client you pick. Only the AI tool configuration differs, covered separately for each client after Step 1.

Step 1: Create an Account and Connect Your Salesforce Org

Every Salesforce MCP server starts the same way: you create an account with the provider, then authorize it against your Salesforce org over OAuth. The screenshots below use Cirra AI, but the flow looks similar wherever you sign up.

1. 1

   Sign up for a free account with your Salesforce MCP server provider — for example, cirra.ai/sign-up.

2. 2

   From its dashboard, choose to add or connect a Salesforce org.

3. 3

   Select Connect via OAuth and log in with your Salesforce System Administrator credentials.

4. 4

   Authorize the connected app when Salesforce prompts for permission.

5. 5

   Your org appears in the dashboard with its name and edition. The connection is live.

What happens at this point

The MCP server reads the org's metadata structure: objects, fields, flows, permission sets, and automation inventory. This context is what the AI tool draws on when you start a conversation about the org.

Step 2: Connect Claude to Salesforce via the MCP Server

Below are the steps to connect Claude to Salesforce.

Install Claude Desktop

If Claude Desktop is not already installed, download it from claude.ai/download. It is available for macOS and Windows. No additional configuration is required after installation before the next step.

Add the MCP server

This is the same process for any Salesforce MCP server — you just need the name and server URL your provider gives you. The example below uses Cirra AI.

1. 1

   Open Claude Desktop.

2. 2

   Open Claude Desktop Settings (on macOS: Claude → Settings, on Windows: File → Settings) and navigate to the Connectors section.

3. 3

   Click “Add custom connector.”

4. 4

   Enter a name and the server URL your Salesforce MCP server gives you (for Cirra AI, that's https://mcp.cirra.ai/mcp), and confirm you trust the connector.

5. 5

   Click Add, then Connect to authorize the connection.

6. 6

   Quit Claude Desktop completely and relaunch it.

7. 7

   Open a new conversation and look for the tools icon (a hammer icon) in the input bar. If your connector appears in the tools list, the connection is live.

Setup takes under a minute. You may need to restart Claude Desktop for the connector to appear after adding it.

Test the connection

Paste this prompt into Claude to confirm the connection is working:

"List the custom objects in my Salesforce org."

If Claude returns a list of the org's custom objects, the connection is active. If it returns a generic answer instead of org-specific data, the MCP connection did not initialize — start a new conversation and try again.

What you can do with Cirra AI once connected

Four specific tasks you can hand to Claude immediately after setup:

• “Create a custom field called Priority on the Case object, type picklist, values: High, Medium, Low” — Claude creates the field, shows the configuration, and waits for your confirmation before deploying.
• “Who has edit access to the Opportunity object? Trace through all profiles and permission sets.” — Claude returns a complete access map tracing every profile, permission set, and group.
• “Review all my record-triggered flows and flag anything that looks risky, like missing fault paths or unbounded loops” — Claude reviews each flow and lists what needs attention.
• “Generate documentation for my Account object including all fields and their descriptions” — Claude produces full plain-English documentation for the object, ready to share or include in a deliverable.

See the full list of tasks Cirra AI can handle on the product page.

Setting Up with ChatGPT Instead of Claude

The account and Salesforce org connection above is identical for ChatGPT. Only the AI tool configuration differs. Below are the steps to connect ChatGPT to Salesforce. If your MCP server has a listing in the ChatGPT app directory (Cirra AI does), that's the fastest path:

1. 1

   Go to chatgpt.com/apps and search for your provider — for example, Cirra AI.

2. 2

   Click Connect on its app listing, then sign in and authorize when prompted.

3. 3

   Start a new conversation, open the app, and test with: “Show me the last 10 Accounts created.”

MCP tool support in ChatGPT requires a ChatGPT Plus or Pro subscription.

Using ChatGPT Codex instead

Codex connects to any custom MCP server the same way: open Codex Settings → MCP servers → Connect to a custom MCP, choose Streamable HTTP, and enter the name and URL your provider gives you (for Cirra AI, that's https://mcp.cirra.ai/mcp).

Setting Up with Cursor Instead of Claude

The account and Salesforce org connection above is identical for Cursor. Only the AI tool configuration differs. Below are the steps to connect Cursor to Salesforce; the steps work for any Salesforce MCP server, and the screenshots show Cirra AI.

1. 1

   Open Cursor Settings (Cmd/Ctrl + ,) and go to Tools & MCP.

2. 2

   Click “+ Add New MCP Server” and enter a name and the server URL your provider gives you, choosing HTTP/SSE as the transport.

3. 3

   Save, then complete the Salesforce authorization flow when prompted.

4. 4

   Confirm the server shows a connected status, then start a new chat and test with: “List the custom objects in my Salesforce org.”

Prefer a config file? Add the same details to .cursor/mcp.json instead — both methods produce the same result.

Common Setup Issues and How to Fix Them

Claude Desktop shows no tools after restart

Cause: The Connectors section may not be showing your connector.

Fix: Re-copy the connector's name and URL from your MCP server's dashboard exactly, validate the JSON at jsonlint.com if you're using a config file, then quit Claude Desktop fully (not just close the window) and relaunch.

OAuth authorization fails in Salesforce

Cause: The connected Salesforce user does not have the System Administrator profile.

Fix: Switch to an account with System Administrator access and retry the OAuth flow from your MCP server's dashboard.

AI tool returns generic Salesforce answers rather than org-specific data

Cause: The MCP connection did not initialize in the current conversation.

Fix: Start a fresh conversation. MCP tool connections initialize at the start of a conversation, not mid-session.

Org data appears outdated after recent changes

Cause: Your MCP server may be holding a cached version of the org metadata from the last connection.

Fix: Disconnect and reconnect the org from its dashboard. This triggers a fresh metadata read.

Frequently Asked Questions

Using Cirra AI as Your Salesforce MCP Server

Cirra AI is the Salesforce MCP server used in the examples above. Unlike Salesforce’s own hosted MCP server, the Cirra AI Salesforce Admin MCP Server writes to the metadata layer directly in conversation: creating objects and fields, building Flows, and managing permission sets and page layouts.

Governance is built into every connection: granular, category-level access controls plus an exportable audit trail mean you stay in control of what Cirra AI can touch. If your team manages more than one Salesforce org, a multi-team hierarchy lets you connect and switch between them from a single setup, rather than reconfiguring per project.

A free plan includes 100 credits per month with no credit card required, and works with Claude, ChatGPT, Cursor, or any other MCP-compatible client. Connecting your first org takes under 10 minutes.