Workday goes headless, opening its HR and finance data to outside AI agents

Workday has unveiled a headless platform, enabling developers to create HR and finance agents utilizing Claude Code, Cursor, and similar tools. At the Workday Rising conference in Las Vegas, the company made its Trusted Agent Platform generally available, granting access to its Model Context Protocol (MCP) server, SQL access to live Workday data, and shared runtime for agents. These components are available on models from Anthropic, Google, Meta, Mistral, Nvidia, and OpenAI.

Workday's CTO, Gabe Monroy, emphasized that the platform ensures agents adhere to Workday permissions, approvals, and audit trails, preventing unauthorized access to sensitive data. The headless approach contrasts with traditional methods of connecting agents to HR and finance data through data lakes or service accounts, which bypass access controls.

Developers can access Workday's ecosystem through Agent-Ready Tools, featuring MCP tools, SQL, and shared runtime. The MCP server, part of Workday Extend AI, requires users to sign in with their own Workday credentials, ensuring that agents operate directly under their account permissions. The server currently offers 275 read tools and 54 tools for data modification. Live Data Query, a component of Workday Data Cloud, allows BI tools and agents to perform SQL queries directly against Workday's data.

Three additional components are now generally available: Agent Harness, the shared runtime that passes user identity, org position, and access rights to agents; Developer Agent, which generates Workday agents and apps from natural-language prompts; and Custom Agents, enabling teams to deploy their own agents built with LangChain or Google's Agent Development Kit (ADK) onto Agent Harness. Two more features, Agent Context Service and Grounded Reasoning, are slated for early access in early 2027.

Monroy demonstrated the headless approach, showcasing an agent connected to a custom MCP server that accessed job requisitions beyond the user's authority and altered employee team assignments without approval. In contrast, using Workday's MCP server and Data Cloud, the same queries returned only the data the user was authorized to see.

The CTO noted that headless offers a choice for customers, with Workday's own tools and APIs being optimal for a limited set of use cases requiring HR or finance domain logic. In other cases, he suggested AI-to-AI integrations, where agents hand requests to Workday's agent orchestrator, Sana, which then calls the company's domain agents.