WorkRight MCP Server — Query Your HR Data from AI Assistants

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external tools and data. The WorkRight MCP serverexposes a single tenant's HR data as MCP tools and resources, so an assistant like Claude or Cursor can answer questions about your people, leave, expenses, payroll, books and compliance directly from live data — no copy-pasting, no exports.

It runs over stdio, the standard MCP transport for a locally launched server, and is authenticated with the same per-tenant bearer token used by the WorkRight REST API.

The shipping MCP server is read-only. Every tool only fetches data, scoped to the token's tenant, and the read shapes deliberately exclude sensitive PII (bank account, PAN, salary). This is the honest current state — not a finished write API.

Safe, capability-gated write tools (apply or approve leave, submit or approve an expense, create an invoice, move a candidate) are on the roadmap and will each be audited and tenant-scoped. By design, the most dangerous actions are never exposed to an AI agent: locking a payroll run or accounting period, computing a full-and-final settlement, closing an offboarding cycle, and reading POSH complaints all stay dashboard-only behind interactive step-up authentication.

The MCP server exposes a large, capability-gated tool set. See the complete, filterable list — grouped by module, with the required capability and a read/write badge on every tool — in the MCP Tool Catalog. New to it? The quickstart walks through creating a token and the client config.

• People — list_employees, get_employee and an employee://{id} resource (non-PII profile fields only).
• Leave — list_leave_requests and get_my_leave_balance.
• Expenses — list_expenses with status and category filters.
• Accounting — list_invoices (receivables) and list_bills (payables).
• Payroll — get_payroll_run, which returns only locked runs (pre-lock drafts are excluded).
• POSH compliance — get_compliance_status for a training-compliance summary, plus a policy://posh resource with the active policy text. POSH complaints are never exposed.

Create a token at Settings → API Tokens in the dashboard, then add the server to your MCP client. A typical client config looks like this:

{
  "mcpServers": {
    "workright": {
      "command": "workright-mcp-server",
      "env": {
        "WR_MCP_API_TOKEN": "wr_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

On startup the server validates the token, resolves your tenant and user, and scopes every tool call to that tenant. From then on you can ask your assistant questions like “which employees haven't finished POSH training?” or “show me the latest locked payroll run” and it will call the matching tool.

Each token is bound to one (tenant, user)pair, so an assistant can never reach another company's data. Read DTOs are non-PII by construction, statutory write-once data stays immutable, and revoking the token in the dashboard instantly cuts off access. See the REST API for the same data over HTTP, or the plans.