Using a Dust agent in a conversation

Two surfaces to talk to your agent

An agent configured with the Spendesk MCP tools can be used in two ways:

1. Dust's web chat — the main interface, for an in-depth, multi-turn conversation.
2. @Dust directly in Slack — mention the agent in a channel or DM, no tool-switching needed.

Either way, the phrasing is the same: describe your need in natural language, the agent figures out which MCP tools to call.

Phrasing examples by role

CFO: "Give me a treasury snapshot across all our entities" → list_companies then get_wallet_summary for each entity. "What's our spend exposure over the last 90 days?" → the agent calls the composite tool spendesk_analyze_spend.

AP Manager: "Which invoices are overdue this week?" → spendesk_get_due_invoices returns overdue invoices ranked by urgency.

Controller: "Check for duplicate invoices this month" → get_payables combined with get_chart_of_accounts.

Finance Analyst: "Compare spend by cost center over the last 3 months" → spendesk_analyze_spend grouped by cost center.

Procurement: "Which purchase orders don't have a matching invoice yet?" → get_purchase_orders cross-referenced with get_payables.

Right inside Slack

The whole conversation stays visible in the channel — colleagues can read it back without having to ask again.

What if the answer isn't quite right?

• Rephrase with more business context rather than technical keywords.
• State the time period explicitly — Spendesk data often spans several years.
• If the agent says it can't fully answer, that's likely a real MCP limitation — see the chapter on the agent as a thinking partner.

Once you're comfortable with one-off conversations, the next step is automating recurring questions: Automating a report with Triggers.