Build better voice, WhatsApp and webchat agents from your AI assistant.
Famulor builds AI assistants for phone calls, websites and messaging. This open-source skill teaches your AI client how to design their prompts, configure their tools, test their behavior and work with your Famulor workspace.
Start with a business task. The skill selects the relevant channel guide and current tools, instead of loading hundreds of endpoint descriptions into every conversation.
npx skills add bekservice/Famulor-Skill --skill famulor-skillThen connect your client to https://app.famulor.io/mcp and sign in with OAuth. The skill supplies instructions; MCP supplies authorized workspace access. Prompt drafting works without a connection. Live data and changes require a Famulor account and the relevant plan and permissions.
Installation guide · Worked examples · Use cases · Prompt templates · API index · Releases
| Use case | Try asking | Guide |
|---|---|---|
| Phone receptionist | “Draft a bilingual receptionist that answers from our knowledge base and escalates urgent requests.” | Voice prompting |
| Appointment booking | “Review my booking assistant. Make it check availability, confirm the time zone and verify the booking result.” | Assistant design |
| WhatsApp support | “Adapt our support assistant for WhatsApp, including human takeover and the messaging window.” | Messaging |
| Website chat | “Create a webchat prompt with readable links and no spoken filler phrases.” | Webchat |
| Email replies | “Draft email instructions with a clear answer and one next step.” | |
| Quality review | “Review these conversations, identify recurring failures and propose regression tests.” | Testing |
| API integration | “Find the REST endpoint, CLI command and TypeScript method to list my assistants.” | API and SDK |
The worked examples turn familiar Famulor demo scenarios into reviewable assistant designs:
| Example | What good looks like |
|---|---|
| Hotel front desk | Clear greeting, hotel knowledge, returned availability, confirmed reservation and recovery |
| Drive-thru | Corrections update one basket; the current quote is confirmed and submitted once |
| Order support | Verified order facts and controlled changes, with WhatsApp, webchat and email adaptations |
| Lead qualification | Useful discovery, an appropriate next step and respectful opt-out handling |
Each includes a filled business brief, complete prompt, setting placement, required runtime capabilities, an illustrative conversation and acceptance checks. The gallery's fictional business role-play is inspiration; the examples explicitly identify the real integrations needed before making customer-facing action claims. Use the deliverable format to see how an agent should present the finished proposal and its verification status.
| Client | Setup |
|---|---|
| Claude Code | /plugin marketplace add bekservice/Famulor-Skill, then /plugin install famulor@famulor; authenticate in /mcp |
| Claude web / desktop | Upload the release's famulor-skill.zip in Customize → Skills, then connect Famulor separately |
| ChatGPT | Install the Famulor plugin for assistant/history review; use a separate custom MCP connection for broader authorized actions |
| Codex | Install the skill, run codex mcp add famulor --url https://app.famulor.io/mcp, then codex mcp login famulor |
| Cursor | Install the skill and add the remote server in MCP settings |
| Gemini CLI | gemini extensions install https://github.com/bekservice/Famulor-Skill |
| Other Agent Skills clients | Use the universal installer above and configure Streamable HTTP MCP with OAuth |
See the step-by-step guide for prerequisites, exact commands, first-run verification, updates and troubleshooting. Client controls depend on the installed version and workspace policy.
The public directory connection is read-only. Its 11 tools review assistants and conversation history. Installing the full skill does not turn that profile into a write-enabled connection. A custom connection has its own approved permissions.
- Channel-specific assistant design, voice prompting, messaging and evaluation guides grounded in Famulor's implementation.
- Adaptable prompts for reception, booking, qualification, WhatsApp, website chat and email.
- Four complete demo-inspired assistant designs, with expected conversations, failure paths and a model deliverable.
- 518 MCP tools in 14 groups and 512 public API operations with REST, CLI and TypeScript SDK navigation, reviewed on 4 November 2026.
- An offline operation lookup, deterministic catalog generation and package validation.
- Portable Agent Skills plus Claude Code, Cursor and Gemini packaging; a separate restricted Claude Store package.
- Versioned ZIP and
.skilldownloads with SHA-256 checksums.
The catalogs are dated navigation snapshots. Live MCP schemas, the public OpenAPI contract and the installed CLI/SDK remain authoritative. Features can depend on your role, plan, channel and enabled integrations. This release does not deploy or change the Famulor service.
- Discover: identify the workspace, channel, current configuration and allowed tools.
- Design: separate instructions from knowledge, tool configuration and routing.
- Validate: test conversational behavior, tool outcomes and the actual target channel.
- Verify: report what was drafted, saved, tested or delivered, with evidence.
The skill preserves the scope of your request. Reviewing a campaign does not start it; drafting a message does not send it. Authorized actions proceed without repeated confirmation, and uncertain writes are checked before retrying.
Found an outdated endpoint or a reproducible prompt failure? Open an issue with a redacted example. See CONTRIBUTING.md for validation and SECURITY.md for private reports.
python3 scripts/validate.py
python3 -m unittest discover -s tests -v
python3 scripts/package-release.py --checkIf the skill helps, a star makes it easier for other Famulor builders to find. Share useful examples and improvements rather than private customer transcripts.