Use cases

Use cases

Three common shapes for using Rutter Connect. Pick the one closest to what you're building before clicking into the Quickstart.

1. Building a finance agent

You're shipping an autonomous (or semi-autonomous) agent that reasons over a company's books and posts entries on its behalf. The agent calls Rutter Connect tools the same way it calls any other MCP server.

What you wire up

• A hosted MCP client (your own backend, or a frontend like Claude Desktop / Claude Code).
• Auth: one Rutter Connect bearer token per agent surface, with merchant connections pre-bound.
• Optional: [[ACTION_CONFIRM]]-style sentinel UI so writes pause for user approval.

What you get from Rutter Connect

• Read tools (list_*, get_*, analytics).
• Write tools (create_*, update_*, delete_*) with destructiveHint annotations.
• Domain-bound dual-slot session so the agent can mix accounting + commerce in one turn.

Watch out for

• Hard-coded domain assumptions in the agent prompt — let get_active_connection drive what's bindable.
• Long-running flows: cache analytics results within a turn (get_13_week_forecast is multi-page).

2. Embedded copilot for merchants

You already have a SaaS dashboard and want to drop an in-product copilot that answers questions like "what's my cash runway?" or "post this receipt as a bill". The copilot reads the merchant's existing Rutter connections.

What you wire up

• A chat panel in your dashboard pointing at Rutter Connect (Streamable HTTP transport).
• A proxy step that mints a short-lived bind token tying the chat session to the logged-in merchant.
• Tool-call rendering in your UI: animated pill while a tool runs, expandable JSON preview of every result, inline write-confirmation buttons.

What you get from Rutter Connect

• Pre-bound connections — the merchant already linked QuickBooks / Shopify in your dashboard, so the copilot doesn't open a Link flow mid-chat.
• Differentiated analytics tools (get_cash_position, get_ar_aging_summary, get_ap_optimization) that compose multiple endpoints into one answer.
• source_url deep links on write responses so users can jump to QBO / Xero to verify what posted.

Watch out for

• Cross-tenant leaks: the bind token must be scoped to the merchant; never reuse session cookies.
• Streaming: render text + tool_use + tool_result blocks inline as Anthropic streams them, ordered by index.

3. Internal ops / finance assistant

You're a finance or ops team with multiple platform connections (QuickBooks for accounting, Shopify for commerce, Amazon for marketplace) and you want a chat assistant that answers cross-platform questions or posts journal entries on demand.

What you wire up

• Claude Desktop or any MCP-compatible client pointed at Rutter Connect.
• Sysadmin role (rutter-sysadmin) for read-only org-wide visibility into connection health.
• Audit log export to your data warehouse so finance has a paper trail.

What you get from Rutter Connect

• Cross-domain answers in one turn ("compare last month's Shopify gross sales to revenue posted in QuickBooks").
• Write confirmation gates so an accidental "delete that invoice" from chat doesn't actually delete the invoice.
• Structured operation_not_supported errors when a platform doesn't allow a mutation, with a suggestion field telling the agent what to do instead.

Watch out for

• NetSuite posted JEs cannot be deleted — Rutter Connect returns a structured error with a "reverse the entry" suggestion.
• delete_invoice voids on most platforms; that's intentional (audit trail), not a bug.

4. Local development

You're prototyping. You want to run Rutter Connect against a sandbox merchant on your laptop without touching production.

What you wire up

• A sandbox connection in the Rutter dashboard (e.g. QuickBooks Online Sandbox).
• Claude Desktop pointed at the dev MCP endpoint.
• RUTTER_CLIENT_ID / RUTTER_CLIENT_SECRET from the dashboard's API Keys page.

What you get from Rutter Connect

• Same tool catalog as production.
• The full set of analytics tools backed by sandbox data.
• No rate limit concerns — sandbox is generous.

Watch out for

• Sandbox data is intentionally distressed (lots of past-due bills); some analytics tools (get_ap_optimization) detect this and return a demo_note flag.
• The next_step hint on login_to_rutter is your friend — it tells you which slot to fill next.

Comparison

What's next

• Quickstart — go from zero to a tool call in five minutes.
• How it works — read the architecture before the security review.