MCP prompt guide + skills

Overview

The Frisbii MCP Connector is only as useful as the prompts you provide it. This guide shows how to structure a prompt so your assistant investigates before it acts, introduces three ready-made skills for the most common billing tasks, and includes example prompts for everything else.

How to write an effective prompt

The prompts that get the best results from the Frisbii MCP Connector generally follow the same shape:

1. Identify the customer by name, email, or ID. If your prompt is ambiguous, ask the assistant to list the matching candidates rather than guess.

2. State the goal in plain language — the complaint, the change you want, or the question you're asking.

3. Set your own guardrails in the prompt where it matters, for example a refund limit, a maximum discount, or "don't cancel, just tell me the options." The connector doesn't enforce business policy for you — the assistant only applies the limits you give it, or the defaults built into a skill (see below).

4. Ask for a preview first on anything that changes data — "show me what you'd do before doing it." This is optional but recommended, especially the first few times you try a new type of request.

5. Ask for confirmation of the result — "verify it went through" — so the assistant reads back the updated state instead of assuming the action succeeded.

Packaged skills

A skill is a ready-made command that packages a full workflow — including guardrails and a built-in verification step — behind a single instruction. Instead of describing the whole process yourself, you invoke the skill by name and supply a few arguments; the assistant handles diagnosis, proposes a fix, waits for your confirmation, and verifies the result before summarizing what happened.

/resolve-billing-issue

The support workhorse: takes a customer and a complaint in natural language, diagnoses the account, proposes the policy-compliant fix, executes it only after your confirmation, verifies the result, and produces a ticket-ready summary.

Example default policy (state your own in the prompt to override):

• Refund cap: €500 — above this, the skill escalates to a human instead of acting

• Refund age limit: 90 days — older charges need your approval first

• Goodwill coupon cap: 20%

How it works:

1. Identifies the customer: If more than one match is found, it lists the candidates rather than guessing.

2. Pulls the billing overview and classifies the complaint: duplicate charge, wrong invoice, charged after cancellation, payment failure, or a goodwill request.

3. Checks the fix against policy. If it exceeds the refund cap or age limit, or the diagnosis is ambiguous, it stops and gives you a diagnosis plus an escalation note instead of acting.

4. Proposes the exact fix, amount, and expected end state, and waits for your explicit confirmation before doing anything.

5. Executes the single agreed fix, then reads back the result to verify it landed.

6. Produces a ticket-ready summary.

Example interaction:

You: /resolve-billing-issue customer=acme-gmbh complaint="we were charged twice in July"

Assistant: Found ACME GmbH. Diagnosis: two charges on 3 July, 41 seconds apart, both against the same invoice — the second is a duplicate. Proposed fix: refund the duplicate charge, €249.00. Within the €500 cap. Proceed?

You: Yes

Assistant: Refunded €249.00. Verified: charge shows as refunded, credit note created, balance €0.00.
— Ticket summary: duplicate charge on the 3 July invoice, refunded in full, resolved. —

/save-customer

A retention flow for a customer who wants to cancel or is at risk of churning. It walks a save ladder — fix a problem first, then pause, then discount — and shows the billing impact before anything is applied.

Example default policy (state your own in the prompt to override):

• Maximum discount: 25%

• Maximum discount duration: 6 months

• Maximum pause duration: 3 months

• Save ladder order: fix an issue → pause → discount (a discount is never offered first if a cheaper rung fits)

How it works:

1. Builds a picture of the account: subscription, payment history, dunning status, and any existing discounts.

2. Classifies the churn driver: a billing problem, a temporary need gap, price sensitivity, or product dissatisfaction, and matches it to the cheapest rung of the ladder that fits.

3. If the account already has an active save discount, it doesn't stack another, it escalates to a human instead.

4. Shows the before/after: current price, price under the offer, duration, and what happens when the offer expires.

5. Applies the agreed offer only after your confirmation, then verifies it's attached correctly.

6. Summarizes the outcome with a follow-up date to calendar.

Example interaction:

You: /save-customer customer=nordwind-media reason="too expensive"

Assistant: Nordwind Media: €890/month, 26 months as a customer, no active discounts, no billing friction. Since this is a price objection, recommending 20% off for 6 months — €890 → €712/month, reverting on 2027-02-18. Within policy. Apply this offer?

You: Yes

Assistant: Applied. Verified: discount attached, next invoice previews at €712.00. Calendar a check-in for 2027-02-18.

/change-subscription

Handles upgrades, downgrades, add-on changes, pauses, resumes, and cancellations — always previewing the entitlement and billing impact first.

Example default policy (state your own in the prompt to override):

• Cancellations require a stated reason

• Downgrades always list every entitlement the customer would lose, before asking for confirmation

How it works:

1. Loads the current subscription, plan, add-ons, and any open balance.

2. Resolves the target — the new plan, add-on, or the pause/resume/cancel operation requested.

3. Builds a preview: the entitlement diff (features gained and lost), the billing impact (old vs. new price, proration, next invoice), and the timing.

4. Waits for your explicit confirmation — cancellations always require a separate, explicit confirmation naming the subscription, never bundled with another change.

5. Executes the single agreed change, then verifies the resulting plan, state, and add-ons match the preview.

6. Summarizes the before/after.

Example interaction:

You: /change-subscription customer=acme-gmbh change="downgrade from Pro to Standard"

Assistant: ACME GmbH is on Pro (€249/month, renews 1 September), with one add-on. Standard is €99/month. Downgrading loses priority support, the higher API rate tier, and SSO; core features and reporting are kept. Billing: €299 → €149/month from the next renewal, no proration. Worth flagging the SSO loss to the customer. Proceed?

You: Yes

Assistant: Updated, effective at the next renewal. Verified: plan shows Standard, add-on intact, no proration invoice generated, as previewed.

More things you can ask

Not every task needs a packaged skill. Here are example prompts for other common jobs.

Customer support

Dunning and revenue recovery

Finance and accounting

Product, pricing, and RevOps

Sales and onboarding