Connecting Salesforce with Claude Through a Governed MCP Integration

OVERVIEW

Vertex delivered a proof of concept connecting Claude to live Salesforce CRM data through the Headless 360 MCP server.

Sales teams could ask questions about CRM data in plain language while Salesforce object permissions, field-level security, and sharing rules remained in effect.

Every write required explicit user approval, and completed changes were re-read to verify the update. Actions were logged in Salesforce under the signed-in user.

OBJECTIVES

• Enable plain-language access to live Salesforce data
• Preserve existing Salesforce access controls
• Validate that Claude’s figures match Salesforce
• Require user approval before CRM changes
• Verify updates and retain Salesforce audit trails
• Test safely before broader rollout

DELIVERABLE

A proof of concept integrating Claude with Salesforce through the Headless 360 MCP server, with user-level authentication, governed read access, approval-based writes, and record verification.

KEY FEATURES

Security That Follows Salesforce Access

Claude connects as the signed-in Salesforce user, so existing object permissions, field-level security, and sharing rules determine which records and fields are available. Actions are logged in Salesforce under that user.

Approval-Based CRM Updates

Before making a change, Claude retrieves the record and displays the current and proposed values. The user reviews and approves the change, Claude updates only the approved fields, and then re-reads the record to verify the result.

Plain-Language Sales Insights

The proof of concept tested requests involving revenue, sales targets, pipeline, opportunities, and team performance. The Sales Cloud plugin also provides skills for daily briefings, pipeline reviews, deal signals, call preparation, and follow-ups.

Controlled Pilot Setup

The integration was limited to selected users through Salesforce permissions and Claude user groups. A sandbox was used to test access and approval behavior before considering production use.

BUSINESS RESULTS

• Accuracy | Claude matched Salesforce figures in the proof of concept
• Access Control | Unauthorized data requests were blocked
• Governed Updates | No CRM changes were made without user approval
• Verification | Approved changes were re-read and confirmed in Salesforce
• Simple Setup | 8 setup steps and one sign-in per user

The proof of concept validated access control, data accuracy, and approval-based CRM updates. Broader rollout and business impact measurement remain next steps.

TECHNOLOGIES

• AI Assistant: Claude Enterprise
• CRM: Salesforce Sales Cloud
• Integration: Headless 360 MCP server
• Authentication: OAuth
• Governance: Salesforce object permissions, field-level security, sharing rules, audit trail
• Deployment Approach: Sandbox validation and controlled pilot rollout

NEXT STEPS

• Expand the pilot to selected reps and managers
• Measure call-preparation time, pipeline-review duration, and activity-logging rates
• Introduce core Sales Cloud skills into daily workflows
• Improve CRM data quality through Salesforce hygiene checks
• Move from sandbox to production after confirming behavior at scale

Download PDF