Life Sciences Cloud Go-Live APIs as Agent Tools: Domain MCP Servers and Agent Fabric Objective: Make the integration work done for go-live the tool layer for agents. The HCP, patient, medical, and order APIs already exist and are governed; wrapping them as domain MCP servers and registering them in Agent Fabric lets Agentforce in LSC and external agents use them with the same policies, consent checks, and audit as any other caller. What's Getting Built HCP MCP server (MCP server) — search and read HCP profiles, affiliations, and consent from the go-live APIs · TOOL search_hcp, TOOL get_hcp Patient MCP server (MCP server) — read enrollment and therapy status for case teams · TOOL get_case_status Medical MCP server (MCP server) — submit medical inquiries and read approved responses · TOOL submit_inquiry, TOOL report_adverse_event Orders MCP server (MCP server) — check order, sample, and return status in SAP · TOOL get_order_status Agent Fabric (governance) — Registry, Broker, Omni Gateway policies, and Visualizer over every agent and tool · REGISTER mcp-server, ROUTE agent-to-agent Value chain phase Phase 06 — Commercial & Patient Services. Marketing to healthcare providers (HCPs) and patient support program delivery. Scope of this plan: Exposing Life Sciences Cloud go-live APIs to agents through domain MCP servers and Agent Fabric. Data in this phase: CRM (Veeva / Salesforce Life Sciences Cloud), prescription data, and logs. Receives from: Phase 05 — Supply Chain & Logistics. Hands off to: Nothing downstream — this phase closes the loop back to Research & Discovery with real-world evidence. Prerequisites Anypoint Platform org with API Manager, Runtime Manager, and a non-production deployment target. Salesforce Life Sciences Cloud org with a connected app and an integration user scoped to the objects this plan writes. Secrets Manager or secure properties configured for every system credential. Named owner for each source system — integration stalls on access, not on code. The go-live System and Process APIs in API Manager — agents consume them, they do not replace them. Integrations deployed from MuleSoft Direct, such as Provider Services for OneKey search and PSP Benefits Verification, count as go-live APIs too and can sit behind the HCP and Patient MCP servers. Agent Fabric enabled: Agent Registry, Agent Broker, Omni Gateway, and Agent Visualizer. Agentforce in LSC, plus any external agent platform such as AWS Bedrock. Response boundaries from medical, legal, and regulatory, and the AE escalation rule. Architecture 1. One MCP server per domain Group tools by domain — HCP, Patient, Medical, Orders — so each server has one owner and one access policy. Generate tools from the existing API specs and describe each one for the model; a vague description is a misused tool. 2. Register and broker Register every MCP server and agent in the Agent Registry with owner, purpose, and data classification. Route agent-to-agent calls, such as Agentforce to a Bedrock agent, through the Agent Broker rather than direct. 3. Least privilege at the gateway Give each agent only the tools it needs, enforced by Omni Gateway policy, not by prompt. Apply the consent policy to every tool that contacts or reads about a person. Require human approval on regulated decisions, and route any adverse event to safety immediately. 4. Trace everything Use the Agent Visualizer to see which agent called which tool with what result. Keep the trace as the audit record for agent actions in regulated processes. 5. Security and GxP compliance Validate the integration under GAMP 5 — requirements, design, IQ/OQ/PQ evidence, and a traceability matrix retained for inspection. Meet 21 CFR Part 11 and EU Annex 11 for electronic records and signatures on every GxP-relevant flow. Preserve ALCOA+ data integrity — attributable, legible, contemporaneous, original, and accurate — with audit trails no operator can edit. Encrypt regulated data in transit and at rest, and keep production payload visibility to validated support roles only. Treat a mapping change on a GxP flow as a controlled change under change management, never as a hotfix. Log every agent tool call with agent identity, tool, inputs classification, and outcome. For pre-built API specifications, implementation templates, and reference architecture, see the MuleSoft Accelerator for Life Sciences on Anypoint Exchange. Testing and validation Call a tool an agent is not granted and confirm the gateway denies it. Ask the medical agent a question containing an AE and confirm safety receives a report. Route an Agentforce request to a Bedrock agent and confirm the Visualizer shows the full path. Phased rollout Phase 1 — Connect Wrap the HCP APIs as an MCP server and register it. Grant one Agentforce agent read-only access. Phase 2 — Map and validate Add Patient, Medical, and Orders servers with least-privilege policies. Broker one external Bedrock agent. Phase 3 — Production Open to more agents under the registry review. Hand trace monitoring to the platform team. Success metrics Agent tool calls passing through the gateway, at 100%. Go-live APIs reused as tools versus new integrations built for agents. AEs from agent conversations reported inside the window. Troubleshooting Agents pick the wrong tool — Likely cause: tool descriptions copied from API summaries. Fix: Rewrite descriptions for when to use each tool and what it returns. Deliverable: HCP, Patient, Medical, and Orders MCP servers over the LSC go-live APIs, registered and brokered in Agent Fabric, with least-privilege and consent policies at Omni Gateway and every call traced.