Workday MCP testing
Workday MCP Testing: Permissions, Audit Trails and What to Test
Introduction
Claude can now act on your Workday data. What proves it only does what it should?
On 13 October, Workday and Anthropic announced that Claude and Sana can exchange information and tasks through MCP, an open standard, with Workday’s permissions and controls intact. For security and integration teams, Workday MCP testing becomes a new job: confirming that every request, from any tool, follows the same rules.
This post explains what changed and gives you a test table for permissions, audit trails and approvals. You will also get an example of a team checking an invoice approval before it goes live, with a clear pass criterion for each check.
Key takeaways
Claude and Sana exchange tasks through MCP, so requests from outside tools can now reach Workday data and actions.
Workday says permissions and controls stay intact. Your tests should confirm that for your own tenant.
Test permissions by role, including people who should be blocked.
Check that every action leaves a record you can read.
Sana in Claude is in early access, with general availability expected in late 2026, so plan your tests now.
Why Workday MCP testing matters now
What Workday and Anthropic announced
According to Workday’s announcement on 13 October 2026, people can start work in Claude or in Sana, with their context and controls intact in each. Claude and Sana exchange information and tasks through MCP, so work carries forward instead of starting over.
What can happen inside Claude
Employees can reach Sana agents, analyze Workday HR and finance data, and take action from within Claude. Workday’s own example is a finance leader who checks invoice exceptions against purchase orders, then approves them in Claude or routes them to Sana.
Why testing changes
Workday says Sana in Claude brings Sana and Workday’s Agent-Ready Tools to joint customers with Workday’s permissions and controls intact. That is a design promise. Your job is to prove it in your tenant, with your roles, your approval limits and your integrations.
Why the next few weeks matter
The Sana in Claude plugin is available to early access customers and is expected to reach general availability in late 2026. As more tools can call Workday AI agents, more users will ask for access. A short window now lets you write tests before that demand arrives.
Example: testing an invoice approval made in Claude
Illustrative, not a real customer. Imagine a finance systems team at a mid-sized organization preparing for Sana in Claude. Its first use case mirrors Workday’s own example: a finance leader reviews invoice exceptions and approves them from Claude.
Before anyone gets access, the team agrees who may approve exceptions from Claude and writes that rule down. Then it writes six checks and records pass or fail for each.
Check | What the team does | Pass when |
|---|---|---|
1. Permissions by role | A finance leader, an AP clerk and an employee with no finance access ask for the same exceptions | Each person sees only what their role allows |
2. Approval limits | The finance leader approves an invoice above their limit | The request is blocked or routed, nothing posts |
3. Audit record | One invoice is approved from Claude and one from Sana | Both leave a record of who, what and when |
4. Routing | An exception is passed from Claude to Sana | Context carries over with no duplicate action |
5. Failure handling | The connection is interrupted mid-request | No partial update, and the error is clear |
6. After a release | Checks 1 to 5 are run again | Results match the earlier run |
The team begins with Workday security roles, because every other check depends on them. It then runs Workday integration testing on the connection and the invoice flow behind it.
After go-live, the same six checks become a Workday regression testing suite. Wider access waits until each check has an owner and a saved result.
How PeopleFlow helps
The new connection is only as safe as the checks behind it. PeopleFlow provides 1000+ pre-built tests for HCM, Payroll, Finance and Integrations, so you can check the roles, approvals and integration flows around MCP through no-code Workday testing, without writing scripts. Explore our Workday testing services and our AI agents for Workday: Configuration, Report, Support, Testing and Integration. As a Workday Innovation Silver Partner, we help you connect new tools with proof, not hope.
Frequently asked questions
What is MCP in Workday?
MCP, the Model Context Protocol, is an open standard that lets AI tools exchange information and tasks with other systems. Workday and Anthropic say Claude and Sana use MCP to work together, so a person can start work in either tool and carry it forward with Workday’s permissions and controls intact.
Is Sana in Claude available now?
Workday says the Sana in Claude plugin is available to early access customers and is expected to be generally available in late 2026. Dates can change, so confirm the current status on Workday’s newsroom or with your Workday account team before you plan a rollout.
How do you test Workday MCP permissions?
Run the same request as users in different roles, including people who should be blocked, and compare each result with what that role may see and do in Workday. Then confirm that every action is recorded, and keep the results as evidence for security and audit teams.
Does Workday’s MCP access replace testing?
No. Workday says permissions and controls stay intact, but your tenant has its own roles, approvals and integrations. Testing proves those rules still hold when a request comes from Claude or another tool, and it should be rerun after every Workday release.
Conclusion
Workday MCP testing turns a promise into proof. Before more people ask for Claude access, check permissions by role, approval limits and audit records, and keep the evidence. Then rerun the same tests after every release. Teams that prepare now can say yes to new tools without losing control of their data. Start with one role and one approval this week.
