By Workday opened the vault on October 13. At its Rising 2026 conference, the HR and finance platform used by more than 10,000 enterprises announced its Trusted Agent Platform is now headless — developers can connect any MCP-compatible AI agent directly to live payroll, recruiting, and financial data without building a custom integration. The MCP catalog ships with 275 read tools and 54 write and delete tools, and the whole thing is available today.
This is a meaningful shift. Before October 13, connecting an AI agent to Workday data meant weeks of custom REST or SOAP integration, per-tenant OAuth configuration, and data sync pipelines that turned live data into stale copies. Now the tools are pre-defined, the permissions are already in place, and developers can start building in Claude Code, Cursor, or VS Code against live data in hours.
What’s Available Now
Four platform components are generally available today. The Agent Harness is the runtime layer that enforces memory, guardrails against prompt injection, sovereign routing for regional data boundaries, model routing (Claude, Gemini, GPT, Llama, Mistral, NVIDIA Nemotron), and built-in evals and traces. The Agent-Ready Tools are Workday’s native MCP server — 275 read and 54 write/delete tools covering Core HR, Financials, Workforce Management, Payroll, Recruiting, and Talent. The Workday Data Cloud gives agents SQL access to live Workday business objects with no data duplication — no stale copies, no security drift. And the Developer Agent converts a plain-language description into a production agent inside your existing IDE.
LangChain and LangGraph are supported. NVIDIA Nemotron is GA. Google ADK is design-partner-only for now, with early 2027 targeted for broader availability. The full MCP tool catalog is available at developer.workday.com with no login required.
What You Can Build Today
Three use cases worth building now. First, a finance agent that validates invoices against open purchase orders in real-time — no manual cross-referencing, no waiting for a nightly data export. Second, an HR self-service agent that answers “Who is Alex’s manager?” or “What is the PTO balance for the engineering team?” from live Workday data instead of a stale spreadsheet. Third, a headcount planning agent that compares approved budgets against open roles the moment a requisition is created.
Workday’s own Talent Acquisition Agent autopilot — which monitors open roles, sources candidates, and schedules interviews — is in early access. The Payroll Agent autopilot, which validates pay data continuously, is early 2027. You can build a custom version of either today using the MCP tools that are already GA. The full Rising 2026 announcement timeline breaks down exactly what ships and when.
The Security Reality Check
Workday’s pitch is clean: agent actions run under the requesting user’s existing permissions. No new permission model to configure, no side doors. Workday CTO Gabe Monroy put it plainly at Rising: “Headless can’t mean lawless. The rules for HR and finance already live in Workday, so customers don’t have to rebuild them.”
That is mostly true. Permissions scope correctly — an agent acting as a junior HR analyst cannot pull compensation data the analyst cannot see in the native interface. But permissions govern what data an agent can access, not what it did access. Standard MCP audit logging records that an agent ran — not which employee record was queried, what arguments were passed, or what was returned. Worker profiles contain the densest PII in the enterprise: social security numbers, compensation history, bank details, dates of birth. That data flows into the agent’s context window without an inspection layer by default.
Workday acquired agentic identity platform C1.ai on the same day as the Rising announcements — a clear signal that the governance gap exists and needs addressing. The acquisition closes early 2027. The Agent Passport feature, which issues verifiable compliance stamps for agent actions, is GA late 2026. Until then, log every MCP tool call yourself: which tool, which user delegated, what arguments, what data classes were returned.
What Is Not GA Yet
Two features matter most for production workloads. Agent Context Service enables agents to retain shared context across sessions and pipelines — early access late 2026, GA early 2027. Grounded Reasoning validates that an agent’s planned action complies with business policy before executing — early access early 2027. If your use case requires either feature, plan around those timelines. The Accounting Agent autopilot and Payroll Agent autopilot are both early 2027 as well.
The Bigger Signal
Workday is the first major HR and finance SaaS to go fully headless via MCP. The pattern it sets matters: instead of rebuilding security rules for agents, enterprises connect existing data governance to the MCP layer. Docusign went MCP-GA in September. AWS Lambda now handles agent invocation without adapters. Claude ships OpenTelemetry tracing in beta today. The enterprise agent stack is assembling one piece at a time — and Workday just handed developers access to the largest single dataset in the enterprise: who works there, what they earn, and how the money moves.
