The short answer: Workday’s own route is Agent-Ready Tools, the MCP tools Workday describes as generally available to all customers through Workday Extend AI. You register an OAuth client, add Workday’s MCP endpoint as a custom connector in Claude or ChatGPT, and the agent works under the signed-in person’s Workday permissions. Do it for one person, in a non-production tenant, with read tools only and approvals on, before anything else.
This post was rewritten on 16 November 2026. An earlier version described the Workday route as early access; that is out of date.
What is Workday’s own route, and is it generally available?
Workday’s Agent-Ready Tools page describes them as Workday connectors built on the Model Context Protocol (MCP), with “hundreds” of tools across Core HR, Financials and Workforce Management and new ones shipping continuously. Workday’s announcement of 13 October 2026 says Agent-Ready Tools, Custom Agents and Developer Agent are “generally available in Workday Extend AI”, and that every agent, whether built by Workday, a customer or a partner, “works under the same Workday permissions, approvals, and audit trails.” In June 2026 the same tools were early access with general availability projected for the second half of the year, so that is the change.
Three nearby pieces have a different status:
- Developer Agent, which builds Extend apps from coding tools such as Claude Code, Cursor and Google Antigravity, is described on its Workday page as “Generally Available starting 2026R2+” and as coming through Extend Professional, the SKU name Workday’s 30 October 2026 release notes changed to Extend AI.
- Sana in Claude, a plugin from the Workday and Anthropic partnership, was described on 13 October as available to early access customers and expected to be generally available in late 2026. Workday said Agent-Ready Tools would reach customers through that plugin later in the year.
- The Adaptive Planning MCP server is separate. Its 2026R2 release note lists a preview date of 2026-08-15 and a production date of 2026-10-23 and says it is released “with limited availability”, only under certain Adaptive Planning SKUs, requested through a Workday Customer Care case.
Workday’s Agent-Ready Tools page says you can connect major agent platforms “over MCP—with no code” and names ChatGPT, Gemini and Copilot, with Claude shown in a logo graphic. Its developer page for external agents is more specific: you need OAuth credentials, and any MCP-compatible platform can connect to the same endpoint with the same credentials. Workday publishes no Claude- or ChatGPT-specific steps, so the steps below come from Anthropic and OpenAI.
What do you set up on the Workday side first?
Workday’s page “Connect an External Agent with Workday (Model Context Protocol server)” gives this order:
- Set up the agent system of record (ASOR).
- Discover the available agent-ready MCP tools.
- Register the external agent.
- Configure the external agent.
- Configure security policies for agent skills.
UMSA customers have the feature enabled automatically. MSA customers enable it through Innovation Services first. Workday’s page says to protect the client secret and bearer tokens and to configure the security policies in step 5 before connecting any agent.
The login is OAuth with an authorization code. You call the authorize endpoint with the client ID, redirect URI and response type, sign in, then exchange the code for tokens at the token endpoint. Workday’s page says access tokens expire after 4 hours and refresh tokens after 30 days. To confirm the connection, send a tools/list JSON-RPC request to the MCP endpoint with the bearer token; a 200 response that lists your registered tools means it works.
How do you add Workday as a custom connector in Claude?
These steps are from Anthropic’s help page for custom connectors using remote MCP. They apply on Free (one custom connector), Pro, Max, Ultra, Team and Enterprise plans.
Pro or Max, for yourself:
- Go to Customize > Connectors.
- Click + Add, then Add custom connector.
- Enter a name, then the server URL exactly as the provider documents it, including any path, and click Continue.
- Review the detected authentication settings and click Continue.
- Under Authentication choose Sign in now, Sign in when needed or No sign in. For a Workday connection choose a sign-in option, never No sign in.
- Under OAuth client, use Claude’s published identity unless the server’s documentation says to use your own client. If you use your own, register
https://claude.ai/api/mcp/auth_callbackas the redirect URL. - Click Add. In a chat, click + at the lower left, then Connectors, and switch it on for that conversation.
Team or Enterprise: an Owner adds the connector under Organization settings > Connectors (Add, then Custom, then Web). On Enterprise, a custom role with Manage access to Libraries can also do this. Each member then goes to Customize > Connectors and clicks Connect, so Claude only reaches what that person can reach. A connector that uses a shared API key gives everyone who uses it that key’s access.
The server must be reachable from Anthropic’s IP ranges over the public internet on port 443. Workday hosts its own endpoint, but a self-hosted vendor or custom server behind a VPN will not connect without allowing Anthropic’s addresses.
How do you add Workday as a custom MCP server in ChatGPT?
OpenAI’s help page “Connecting custom MCP servers in ChatGPT” describes custom MCP connections on ChatGPT web for Business and Enterprise/Edu workspaces. The flow it describes at the time of writing:
- Open Plugins, click the plus button and choose Add custom MCP server.
- Enter a name and an optional description.
- Under Connection, enter a public HTTPS MCP endpoint URL, or choose a Secure MCP Tunnel for a local server.
- Configure authentication, accept the risk warning, and select Create as a plugin.
- Review the discovered tools, install the plugin, start a new chat and type @ to select it.
On Business, only admins and owners can create and publish custom MCP plugins. On Enterprise/Edu, admins can let selected members create connections, and only admins and owners publish. OpenAI’s menus have changed more than once, so if yours differ, follow the current version of that page.
What is the checklist for connecting safely?
- Use per-user OAuth, not a shared integration system user. Workday’s external-agent route signs in as the person, and Workday says actions run under that person’s permissions. An ISU is a service account: every call runs with its permissions whoever asked, and it is the wrong default for a chat assistant. The guide on user-delegated OAuth versus ISU lays out the difference, and authorization code grant setup walks through the registration.
- Start in a non-production tenant. Connect to a Sandbox, Implementation or a free personal developer tenant first. Testing Workday MCP connectors on a free developer tenant covers that setup.
- Start read-only. Enable only read tools. In Claude, switch off any tool that writes to Workday in the Search and tools menu. In ChatGPT, use the permission level that reads without asking and asks before changes, and on Enterprise/Edu disable individual write actions.
- Set Workday’s security policies for agent skills before the first call. This is step 5 of Workday’s setup, and the agent can only do what the person’s security groups and those policies allow. Auditing security before enabling AI agents is the checklist for it.
- Keep tool approvals on. Claude asks before it calls a tool; use Allow always only for tools you trust to run unattended. Anthropic warns that Research can call connector tools without further approval, so disable write tools before using it.
- Log and review the calls. Workday records agent activity in the same audit trail as any other action. Read it, and watch the Platform Consumption Console, because each Agent-Ready Tool call is metered in Flex Credits.
- Treat tool changes as a change request. Anthropic notes that server developers can change tool behaviour without notice, and OpenAI describes a frozen tool snapshot until an admin reviews and publishes an update. Re-check the tool list after any change.
- Keep secrets out of chats. The client secret and tokens belong in the connector’s authentication settings only, and Workday’s page says to protect them.
What does it cost to call Workday tools from Claude?
Workday defines a Tool Call as a call to an Agent-Ready Tool “from an agentic Extend app or from an externally managed AI agent”, each tool designated Base or Advanced. The Flex Credits Rate Card dated 5 November 2026 prices them per 100 calls: Base 5 credits for custom agents and 10 for other agents, Advanced 25 and 50. The Rate Card gives no dollar figure, so ask your account team what a credit costs under your agreement.
Where does Kiweely fit?
Kiweely Vine is a desktop app, not a connector for Claude or ChatGPT. It runs on Kiweely Credits or on the ChatGPT or Claude plan or API key you already have. Its AI agent signs in to Workday the way you do and does the work in the tenant itself: configuration and business processes, security, reports and calculated fields, data loads, Studio integrations and Orchestrations, and Extend apps, not only the slice of Workday an API client can see. Each conversation is pinned to one client’s tenants, so consultants who work across clients keep each client’s work separate. For the choice between a Workday MCP server and Claude in Kiweely, read Claude Code with a Workday MCP server or with Kiweely, and see the Security page.
Frequently asked questions
Is there a Workday connector for ChatGPT or Claude?
Workday's Agent-Ready Tools are its MCP tools, and Workday says any agent that supports MCP can use them. Workday's Anthropic partnership also describes a Sana in Claude plugin, which Workday listed as early access with general availability expected in late 2026. Workday publishes no ChatGPT-specific setup steps; ChatGPT connects the same way as any custom MCP server.
Do I need an integration system user to connect Claude or ChatGPT to Workday?
Not for Workday's own route. The Workday page for external agents uses an OAuth authorization-code login, so the agent acts as the person who signs in. A vendor MCP server built on an integration system user runs with that service account's permissions instead, whoever asked.
Do I need Workday Extend to use Agent-Ready Tools?
Workday says Agent-Ready Tools are generally available to all customers through Workday Extend AI, and its external-agent page limits the feature to Extend Professionals, the SKU Workday has since renamed Extend AI. Confirm your package with your Workday account team.
What does it cost when Claude calls Workday tools?
Workday meters Agent-Ready Tool calls in Flex Credits. On the November 5, 2026 Rate Card a base tool call costs 10 credits per 100 calls when used by an agent other than a Workday custom agent, and an advanced call costs 50 credits per 100. Workday's pages do not publish a dollar price per credit.