Salesforce MCP — AI-Native Intelligence Server

The organisation faced a critical "data translation" gap. High-value data was distributed across complex, interlinked Salesforce objects — requiring manual extraction, specialist technical knowledge, and hours of synthesis before any insight could be acted on. This bottleneck blocked real-time, data-driven decision-making across the entire org. The objective was to eliminate that gap entirely by building an AI-native layer directly on top of the Salesforce data model.

• Translated vague stakeholder "insight" requirements into a precise, versioned technical schema before a single line was written
• Designed a metadata-mapping layer to ensure the LLM accurately resolved complex Salesforce relationship fields — preventing hallucinations at the data layer
• Implemented dynamic SOQL query generation so every response reflects live org data, not cached snapshots
• Balanced generative AI flexibility with enterprise-grade security constraints, mapping the MCP server's access scope to existing Salesforce user-level permissions

• Directed a cross-functional team spanning Engineering, DevOps, Product, QA, and Business Development from spec through pilot delivery
• Deployed the Anthropic MCP protocol with Claude Desktop as the client/host — enabling both technical and non-technical staff to query Salesforce in natural language
• Surfaced automated impact-gap analysis, programmatic trend reports, and funding-to-outcome mapping without any manual data export
• Successfully converted the internal proof-of-concept into a repeatable implementation framework now being productised for diverse client environments