Claude for Salesforce: How to Use Anthropic’s AI in Your CRM

Anthropic's Claude has become one of the most capable AI models for business writing, reasoning and multilingual work, and sales teams are asking the obvious question: can I use Claude inside Salesforce? The answer is yes, and you don't need a custom development project to do it.

Why teams choose Claude for CRM work

Claude stands out in exactly the areas that matter for sales: long context (it can take in a lot of account history in one pass), careful, natural-sounding writing for customer-facing emails, and strong performance in European languages, so French, German, Dutch, Spanish or Polish follow-ups read naturally. For teams that care about data handling, Anthropic's commercial API terms state, at the time of writing, that business API data is not used to train models by default. Terms can change, so check the current ones in your own agreement before go-live.

The integration problem

Connecting Claude to Salesforce yourself has a catch whichever way you do it: copy-pasting record data into claude.ai breaks your security model and leaves no audit trail; building a custom Apex integration means callouts, authentication, prompt management and maintenance; and middleware platforms add a monthly bill plus another place your CRM data travels through.

How AI Engine connects Claude to Salesforce

AI Engine is a managed package that runs Claude natively inside Salesforce through a Named Credential, Salesforce's own secure mechanism for external API calls. The setup has four steps:

1. Get an Anthropic API key from console.anthropic.com (pay-as-you-go, billed to you by Anthropic).
2. Create a Named Credential in Salesforce Setup pointing at the Anthropic API, with your key stored in an External Credential: never in code, and not shown to AI Engine users. Allow the CloudConnected namespace to use it and give your users access to the credential. The steps are in Connect your AI provider.
3. Select Anthropic Claude as your provider in the AI Engine admin panel's Org Profile and pick a model.
4. Assign permission sets and add the components: add the Sales Agent to the record pages you choose and to the Utility Bar, and your reps get an AI assistant right there.

From that point, Claude powers record summaries, follow-up email drafts, lead and opportunity scoring with a written reason, meeting prep briefs and natural-language questions about your own CRM data, all without anyone leaving Salesforce.

Your key, your control, your data boundaries

Because AI Engine uses a bring-your-own-key model, the relationship with Anthropic is yours: you control spend, rate limits and which Claude model your data goes to. AI Engine enforces your existing Salesforce security on top: sharing rules and field-level security are respected, an admin decides in Agent Objects which objects and fields the AI works with, personal data is masked by default before it reaches the model, and AI Engine does not store the full text of prompts and answers (its activity log keeps usage metadata, plus a short error message when a call fails). AI usage is billed directly by Anthropic on your own account. Read more about the security model in Secure AI for Salesforce.

Claude, OpenAI or Azure, or all three

AI Engine isn't locked to one vendor. You can run Claude today, switch to OpenAI tomorrow, or use Azure OpenAI for data-residency requirements (web-search features on Azure go to OpenAI; see Azure OpenAI in Salesforce). Once the new provider's Named Credential is set up, switching is a configuration change in the Org Profile, not a migration. A practical start is the provider your company already has an account with; evaluate from there.