MCP ServerHTTP

Salesforce MCP Server

Salesforce’s hosted MCP endpoint at mcp.salesforce.com/mcp lets AI clients use your CRM the same way authenticated integrations do: searching leads, updating opportunities, and summarising pipeline health, subject to your org’s permissions and the token you provide. You typically use a connected-app OAuth access token or a flow-approved token from your Salesforce admin, then run sales questions from chat that would otherwise need SOQL in the dev console or many UI clicks.

Hosted URL

https://mcp.salesforce.com/mcp

Suggested model

Claude Sonnet 4.5

MCP Playground runs 30+ models on the same workflow: switch anytime, or use Compare mode to run several in parallel and balance quality vs. cost.

Auth

Salesforce access token from a Connected App (OAuth) or an approved org session flow. Follow your org’s security policy for client secrets and refresh handling.

What the Salesforce MCP server does

How models use it and what it is built for.

The Salesforce MCP is for revenue and operations teams on Sales Cloud. Models can help reps prepare for calls, managers roll up this quarter’s pipeline, and ops spot stale opportunities — always bounded by the API access on your token. It does not replace your security review: lock down the connected app, use sandbox orgs for experiments, and treat the agent like any high-privilege integration with logging and DLP in place.

Tools the Salesforce MCP server exposes

Typical tools an AI model can call. Exact names vary by version.

  • Query and search leads, contacts, accounts, and opportunities (per object access)
  • Read and update standard fields, tasks, and activities where your token allows
  • Summarise pipeline and stage distribution for a time range or territory

Example prompts to try

Copy any of these into MCP Agent Studio after connecting.

  • List open opportunities over $50k that have had no activity in 21 days.

  • Show new leads this week and which are still unassigned.

  • Summarise the pipeline by stage for the EMEA team this quarter.

Models on MCP Playground

This is not a single-model product: you get the same MCP connection with 30+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.

Default pick for Salesforce

Claude Sonnet 4.5

Claude Sonnet 4.5 handles long CRM context and natural-language to filter mapping. For tight SOQL assist, test GPT-5.2 in Agent Studio on the same prompt.

Try the Salesforce MCP server in your browser

Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 30+ models, and start chatting — no install required.

Try Salesforce in Agent Studio

Salesforce MCP server — FAQ

Common questions about connecting, scoping and using it safely.

What is the Salesforce MCP server?

It is Salesforce’s hosted remote MCP for connecting AI tools to your org’s CRM data through Model Context Protocol tools, similar in scope to other REST-based integrations, but with prompts and tool descriptions tuned for large language models.

What credentials do I use for the Salesforce MCP server?

You need an OAuth access token for an API user or a user completing the OAuth web flow, issued by a connected app with the right API and object permissions. Your admin may need to allow that connected app; never paste a production token into a client you do not trust.

How do I test safely in production?

Start in a full sandbox, use a read-only or narrow-profile integration user, and add write scopes only after you validate behaviour. Enforce field-level security and record sharing so the agent only sees the accounts and opportunities you intend.

Can I use the Salesforce MCP from the browser in MCP Playground?

Yes, if your org allows the token in that environment. Add the server URL and token in Agent Studio and pick any model. For enterprise SSO, follow your org’s device and OAuth policy.

What model is best for Salesforce questions?

Claude Sonnet 4.5 is a strong default for pipeline narratives and long opportunity threads. You can A/B a cheaper open model in Compare mode in Agent Studio for list-heavy, repetitive pulls.

Other MCP servers

More on MCP Playground