The integration uses Model Context Protocol (MCP). Users can start a task in Sana or Claude and continue it in the other interface. Early access is available now. General availability is expected in late 2026.
Technical specifications
| Parameter | Details |
| Platforms | Workday Sana + Claude |
| Protocol | MCP |
| Main use cases | HR, finance |
| Authorization | Workday permissions |
| Audit | Workday audit controls |
| Business rules | Workday policies and workflows |
| User approval | Required before changes |
| Task handoff | Sana ↔ Claude |
| Availability | Early access |
| GA | Late 2026 |
How it works
Two execution paths are supported:
User → Claude → MCP → Sana Agent → Workday
and:
User → Sana → Claude reasoning → Sana → Workday
Claude handles reasoning and natural-language interaction. Sana remains the orchestration layer for Workday agents and business workflows.
Workday remains the system that controls:
- permissions;
- approvals;
- business rules;
- audit history.
Claude does not receive unrestricted access to the Workday tenant.
Whichever way they come in, the rules stay anchored in Workday.
What users can do
From Claude, users can invoke Sana agents for supported HR and finance tasks.
Examples include:
- checking open positions and headcount;
- reviewing budget data;
- finding internal candidates;
- investigating invoice exceptions;
- comparing invoices with purchase orders;
- initiating supported Workday actions.
Every change must be confirmed by the user.
Previous Sana vs. Sana + Claude
| Capability | Previous Sana | Sana + Claude |
| Main interface | Sana | Sana or Claude |
| Workday agents | Yes | Yes |
| Workday permissions | Yes | Yes |
| Audit controls | Yes | Yes |
| Claude reasoning | Separate/limited | Integrated |
| Task continuation | Mainly inside Sana | Sana ↔ Claude |
| MCP | No | Yes |
| Claude as Workday entry point | No | Yes |
The main change is not a new Claude model. It is a new interoperability layer that allows Claude to act as another interface for Workday workflows.
Security model
The integration reuses Workday's existing security model.
That includes:
- role-based permissions;
- business-process security;
- audit logging;
- user confirmation before actions.
Developers do not need to build a separate authorization system specifically for Claude.
What has not been disclosed
Workday and Anthropic have not published:
- Claude model version used for each workflow;
- MCP rate limits;
- latency guarantees;
- API pricing;
- context limits;
- maximum workflow duration;
- full list of supported Workday actions.
Existing Workday-Anthropic relationship
Anthropic uses Workday for finance and HR. Workday uses Claude Enterprise and Claude Code internally, including for product development. This means the partnership is not limited to a third-party connector: both companies already use each other's platforms.
Bottom line
The architecture is straightforward:
- Claude = reasoning and interface
- Sana = agent orchestration
- Workday = permissions, rules, data and transactions
The technical change is MCP-based access to Workday workflows from Claude without replacing Workday's existing governance layer.
