AI Assistant

The AI Assistant page connects Claude to the WFM data warehouse via MCP (Model Context Protocol). It's a setup screen — once configured, you use Claude (web, desktop, or any MCP-aware client) to ask questions about your WFM data in natural language.

What you can ask

Once connected, Claude has tools for:

• list_staff, list_clients, list_jobs, list_tasks, list_timesheets
• get_billability_report(date_from, date_to, staff_uuid?)
• get_personal_metrics, get_team_metrics
• get_overtime_report(year, staff_uuid?) — same data the dashboard's overtime pages show.
• list_quotes, create_quote, update_quote, add_quote_task, update_quote_task — full quote CRUD against the WFM API.

Sample questions:

• "How many billable hours did the team log in March 2026?"
• "What's Dario's overtime balance for 2026?"
• "Which clients have we billed the most against this quarter?"

Setup steps

The page lays them out in order:

1. Copy the MCP gateway URL — points at the AgentCore-hosted MCP runtime.
2. Add the connector in Claude (web or desktop) — paste the URL, give it a name like "WFM".
3. Sign in with Google — Claude's OAuth flow exchanges your @copebit.ch credentials for a Cognito-backed JWT scoped to the MCP runtime.
4. Test the connection — ask Claude "who am I?" — it should reply with your WFM staff profile.

Troubleshooting

• "Tool unavailable" errors — the MCP runtime data is loaded on the first request. Wait 30 s and retry.
• Empty results for "current" data — the warehouse syncs from WFM nightly. Same-day timesheets won't be visible until the next sync.