Anthropic Adds 37 Salesforce Skills to Claude

Anthropic has launched an open beta of Salesforce in Claude, a new integration that embeds 37 prebuilt sales skills directly into the AI assistant. Part of a broader Claudeforce partnership announced alongside Salesforce's Q2 FY27 earnings, the plugin is already being used by roughly 7,000 sellers at companies like GitLab, Siemens, and Legora. The tool allows users to gather account context, evaluate opportunities, update pipeline metrics, and generate forecasts directly from Claude, requiring explicit seller approval before writing any changes back to the CRM.

The integration runs on Salesforce's AIforce enterprise integration layer and utilizes the Model Context Protocol (MCP) alongside Salesforce's Headless 360 architecture. This setup ensures that Claude inherits existing Salesforce user permissions, meaning the AI can only access objects, fields, and records the specific user is already authorized to see. Administrators must request beta access through AgentExchange, and the plugin is available across all paid Claude plans, including Team and Enterprise, where Anthropic does not train its models on customer data by default.

As part of the Claudeforce agreement, the two companies are also engaging in reciprocal internal deployments. Salesforce is adopting Claude Code and Claude Enterprise for its engineering teams, while Anthropic is utilizing Salesforce as its primary CRM. While the initial release focuses strictly on sales workflows, Anthropic plans to roll out additional prebuilt skills for other departments starting in late 2026.

For enterprise practitioners, this release provides a secure blueprint for connecting AI agents to core operational systems without creating redundant permission layers. By consolidating data from Salesforce, Slack, and email, the plugin eliminates the manual chore of cross-referencing multiple applications. However, because the system relies heavily on existing CRM structures, practitioners must ensure their data hygiene is pristine; stale records or inconsistent field definitions will directly degrade the quality of Claude's automated summaries and forecasts.