Salesforce integrates MCP in Agentforce 3 adding tooling, governance, and observability for Enterprise deployment
unveiled June 23, 2025, brings enterprise-grade maturity to AI agents by tackling 2 key challenges: visibility and control. The introduction of Command Center offers a unified observability hub—complete with real-time KPIs, session tracing, AI-driven insights, and native integration into monitoring tools and Service Cloud—for monitoring, measuring, and optimizing agent performance . Native support for the Model Context Protocol (MCP) (plus future A2A support) transforms Agentforce into a plug‑and‑play ecosystem, connecting with tools like AWS, PayPal, Cisco, Box, Google Cloud, Stripe and more via the expanded AgentExchange . Under the hood, an enhanced Atlas architecture delivers lower latency, better accuracy, greater resiliency, and LLM flexibility (including Anthropic Claude now and Google Gemini later) . Real-world traction spans 8 000+ customers, with ROI examples like a 15% reduction in case handle times and 70% autonomous chat resolutions during tax season . In short, Agentforce 3 transforms AI agents from experimental novelties into controllable, scalable, and productive teammates for enterprise workflows.