> ## Documentation Index
> Fetch the complete documentation index at: https://docs.famulor.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Use Cases & Prompts

> Practical use cases for the Famulor MCP server and skill — with ready-to-use prompts and best practices for ChatGPT, Claude, Cursor & co.

# Use Cases & Prompts

The [Famulor MCP server](/en/mcp/server) and the [Famulor Skill](/en/mcp/famulor-skill) let you control the entire platform in plain language. This page shows what that looks like in practice — grouped by use case, with copy-paste prompts.

<Note>
  All prompts work in ChatGPT, Claude, Cursor, and any other connected MCP client. Setup: [MCP Client](/en/mcp/client).
</Note>

## Creating & optimizing assistants

The most common starting point: create, test, and iterate on voice agents — without opening the platform UI.

**Prompts:**

```text theme={null}
Create a German inbound assistant "Lisa" for my hair salon.
Pipeline mode, female voice, GPT-4.1-mini. Tasks: take appointment
requests, quote prices, answer opening-hours questions.
```

```text theme={null}
Show me all available German voices and multimodal models.
```

```text theme={null}
My assistant #123 interrupts customers too often. Disable
interruptions and shorten the initial message to one sentence.
```

```text theme={null}
Enable a conversation-ended webhook for assistant "Lisa"
pointing at https://example.com/hook.
```

**Best practice:** Have the model call `get_voices`, `get_languages`, and `get_models` first instead of guessing IDs — e.g. "Show me the options first, then create…".

## Outbound campaigns & lead management

Set up campaigns, import leads, and control them — at scale.

**Prompts:**

```text theme={null}
Create a campaign "Reactivation Q3" with assistant #45.
Calls Mon–Fri 10am–5pm only, max 2 retry attempts.
```

```text theme={null}
Add these leads to the campaign: [list of name + phone number].
Set the variable "product" to "Premium plan" on all of them.
```

```text theme={null}
Pause all running campaigns — we forgot about a public holiday.
```

```text theme={null}
Which leads in campaign #12 have status "not reached"?
Bump their priority.
```

**Best practice:** Always pass phone numbers in E.164 format (`+49170…`). Lead variables must match the variables defined on the assistant — you can't invent new variable names per lead.

## Call analysis & reporting

Query and analyze call data — the model filters, aggregates, and summarizes.

**Prompts:**

```text theme={null}
Show all calls from today. On which ones was the call goal
not reached, and why?
```

```text theme={null}
Summarize the last 20 calls of assistant "Lisa":
what were the most common customer requests?
```

```text theme={null}
Get the transcript of call #9876 and suggest concrete
improvements to the system prompt.
```

**Best practice:** Combine analysis and action in one prompt: "Analyze this week's calls and adjust the system prompt accordingly — show me the change before saving."

## Knowledge bases (RAG)

Connect knowledge so assistants answer from facts.

**Prompts:**

```text theme={null}
Create a knowledge base "Menu" and ingest
https://my-restaurant.com/menu as a document.
```

```text theme={null}
Attach the knowledge base "Menu" to assistant #88 in
"prompt" mode so it always has the menu at hand.
```

```text theme={null}
Our prices changed — update the "Price list" document
in knowledge base #5.
```

**Best practice:** Use `function_call` mode (default) for on-demand facts, `prompt` mode when the bot needs content permanently (menus, delivery services). For multimodal/dualplex assistants, `function_call` is required.

## Conversations, WhatsApp & SMS

Run chat channels — including human takeover.

**Prompts:**

```text theme={null}
Show all open conversations. Where is a customer waiting
for a reply?
```

```text theme={null}
Take over conversation #321: disable the AI and send
"A colleague will be with you personally in a moment."
```

```text theme={null}
Send the WhatsApp template "appointment_reminder" to +49170…
with tomorrow's 2pm slot.
```

```text theme={null}
Send an SMS to +49171…: "We received your callback request."
```

**Best practice:** Before freeform WhatsApp, have the model check whether the 24h session window is open (`get_whatsapp_session_status`) — outside the window only templates are deliverable.

## Phone numbers & SIP

Search, purchase, and manage numbers.

**Prompts:**

```text theme={null}
Search for a German phone number with the 030 area code and buy it.
```

```text theme={null}
Give the number +4930… the nickname "Support Hotline Berlin".
```

```text theme={null}
Create a SIP trunk for our existing PBX: [credentials].
```

**Best practice:** Number purchases cost money and `release_phone_number` is irreversible — ask for a confirmation step before both.

## Folders & organization

Folders and labels keep agencies and teams organized — especially with many assistants.

**Prompts:**

```text theme={null}
Create a folder "Client: Dr. Meier Dental" and move
assistants #12 and #13 into it.
```

```text theme={null}
Create the labels "Production" (green) and "Test" (yellow)
and tag all assistants accordingly.
```

```text theme={null}
Which assistants have no folder yet? Suggest a sensible structure.
```

**Best practice:** An assistant lives in exactly one folder but can carry multiple labels — use folders for clients/projects, labels for status or environment.

## Mid-call tools & automation

Custom HTTP integrations the assistant calls live during a conversation.

**Prompts:**

```text theme={null}
Create a mid-call tool "check_availability" that calls
GET https://api.example.com/slots and returns free slots to the bot.
```

```text theme={null}
Build a tool that looks up order status by order number,
and attach it to assistant "Support Bot".
```

**Best practice:** Write tool descriptions from the bot's perspective ("Call this tool when the customer asks about…") — they control when the LLM uses the tool.

## Guided onboarding with the Famulor Skill

The [Famulor Skill](/en/mcp/famulor-skill) complements the MCP server with **industry knowledge and a structured 4-phase flow** — ideal when a complete assistant needs to be built for a new client. Runs in Cursor, Claude Code, Codex, OpenCode, and Gemini CLI.

**Prompts:**

```text theme={null}
Onboard a new client: Dr. Meier dental practice in Berlin.
```

```text theme={null}
Create an inbound bot for my restaurant — walk me through
the setup step by step.
```

```text theme={null}
New client, trades industry (emergency plumbing). Suggest the
typical tasks, a post-call schema, and an end_call description.
```

```text theme={null}
Test the assistant "Sofia" via chat: book a table for 4 on
Friday at 7pm and verify all data lands in the post-call schema.
```

```text theme={null}
The client finds the bot too formal. Rewrite the system prompt
in a more casual tone and show me the diff.
```

**Best practice:** Let the skill drive the flow — it asks phase by phase (discovery → tech → industry → system prompt) and proposes the right defaults per niche (voice, ambient sound, guardrails). Always have the system prompt confirmed before deployment.

## End-to-end workflows

The real strength: multiple steps in a single request.

```text theme={null}
Onboard a new client: restaurant "Bella Vista".
1. Create a knowledge base from https://bellavista.de
2. Create an inbound assistant "Sofia" (female, German, pipeline)
   for reservations and menu questions
3. Connect both, create the folder "Bella Vista"
4. Finish with a summary of the configuration
```

```text theme={null}
Weekly report: list all calls from the last 7 days per assistant,
summarize success rates, and name the top 3 reasons goals
weren't reached.
```

## General best practices

| Rule                            | Why                                                                              |
| ------------------------------- | -------------------------------------------------------------------------------- |
| **Read before write**           | "Show me first…" before changes — the model fetches real IDs instead of guessing |
| **Confirm destructive actions** | Ask for explicit confirmation on `delete_*`, campaign stops, and number releases |
| **E.164 everywhere**            | Phone numbers always as `+49…` — other formats fail                              |
| **Preview changes**             | "Show me the new system prompt before you save it"                               |
| **One workflow, one prompt**    | Phrase multi-step jobs as a numbered list — the model works through it in order  |
| **API key via OAuth only**      | Never paste the key into configs or prompts — sign-in happens in the browser     |

## Resources

* [MCP Server](/en/mcp/server) — all 76 tools at a glance
* [MCP Client setup](/en/mcp/client)
* [Famulor Skill](/en/mcp/famulor-skill) — guided onboarding
* [API reference](/en/api-reference/)