MCP is the NL-first surface: the agent should pass a role query + company, not invent std_grade codes. Resolution happens on Arxena’s side.

​
Prerequisites

  1. Create a workspace API key in Settings → APIs & Webhooks
  2. Treat the JWT like a password (Authentication)
Setup overview: arxena.com/solutions/mcp-server

​
Connect Cursor

{
  "mcpServers": {
    "arxena": {
      "url": "https://mcp.arxena.com/mcp",
      "headers": { "X-API-KEY": "<workspace-api-key-jwt>" }
    }
  }
}

​
Connect Claude Desktop

{
  "mcpServers": {
    "arxena": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.arxena.com/mcp",
        "--header",
        "X-API-KEY: ${ARXENA_API_KEY}"
      ],
      "env": {
        "ARXENA_API_KEY": "<workspace-api-key-jwt>"
      }
    }
  }
}
Quit Claude Desktop fully after editing.

​
Prompts

  • “Get me HR leadership at Apple”
  • “Find CHRO-type people at Apple”
  • “Who is Head of HR at Apple?”
Prefer search_people_api with naturalLanguage (e.g. “CEO at StayVista”). Use explicit std filters only when you already have them. Public vocabulary nouns: taxonomy constants. In Ask AI on an open org chart, load org-structure-insights, map with list_org_chart_positions, paint with highlight_org_chart({ nodeKeys }), and drill stored people with get_org_chart_node_people({ nodeKey }) so answers cite node keys plus profile headline / summary. Optional web corroboration: app_exa_web_search and google_serp_search (Ask AI catalog flags IS_SEARCH_EXA_ENABLED / IS_SEARCH_SERP_ENABLED).