Famulor MCP Server

Control your Famulor AI voice agents — calls, leads, campaigns, knowledge bases, conversations, WhatsApp, SIP trunks — directly from ChatGPT, Claude, Cursor, Windsurf, VS Code, and any other MCP-compatible client.

Connect a client

Setup for ChatGPT, Claude, Cursor, and more

Hosted server

app.famulor.de/mcp — no install required

GitHub

Source code and deployment

Overview

The Famulor MCP server exposes the full Famulor platform as 76 tools your AI assistant can call. Ask in plain language — the model chooses the right tools. All tools return structured output (output schemas + structuredContent) and carry title annotations for directory-compatible clients. Example prompts:

“Create a German sales assistant using GPT-4.1-mini and the Susi voice”
“Start campaign #42 and add these 50 leads”
“Show me all calls today where the goal wasn’t reached”
“Create a ‘Clients’ folder and move all inbound assistants into it”

Need inspiration? See Use Cases & Prompts — grouped scenarios with ready-to-use prompts and best practices.

New to MCP? Start with MCP Client setup — add the server URL, then sign in with OAuth and your API key in the browser.

Hosted endpoints

Endpoint	URL
MCP (canonical)	`https://app.famulor.de/mcp`
Base URL	`https://app.famulor.de/mcp`
Health	https://app.famulor.de/health
OAuth metadata	`https://app.famulor.de/.well-known/oauth-authorization-server`

Use https://app.famulor.de/mcp as the MCP URL in connector dialogs.

What you can do

Area	Capabilities
Assistants	Create, update, delete AI voice agents · list voices, languages, LLM/multimodal models, TTS/STT providers · toggle inbound and conversation-ended webhooks · organize via `folder_id` / `label_ids`
Folders & Labels	Organize assistants into folders (one folder per assistant) · tag with colored labels (multiple per assistant) · create, rename, delete
Calls	Make outbound calls · list / get / delete call records
Campaigns	Create campaigns, retry rules, allowed-hours windows · start, stop, delete
Leads	Add leads (incl. secondary contacts) · update status and merge variables · delete
Knowledge bases	Create knowledge bases · upload website-scraped documents · update / delete
Phone numbers	Search by country and dial pattern · purchase · set nickname (`update_phone_number`) · release · SIP trunk CRUD
Conversations	List, read, reply to chats · enable/disable AI per conversation (human takeover)
WhatsApp	List senders and templates · check 24h session window · send template or freeform messages
SMS	Send via your purchased numbers
Mid-call tools	Register custom HTTP integrations the assistant can call mid-conversation
AI Replies	Generate context-aware replies for external chat platforms

Authentication

The server uses MCP OAuth 2.1 (RFC 8414 metadata, RFC 7591 Dynamic Client Registration, PKCE S256):

Your MCP client opens a browser tab.
You paste your Famulor API key.
The client receives an access token; the server encrypts it with AES-256-GCM.

Your API key never lands in any database — it lives only inside the token your client stores. Tokens expire after 30 days; your client re-prompts for sign-in.

Do not put your API key in mcp.json, environment variables in shared configs, or version control. Use OAuth sign-in through your client.

Supported clients

ChatGPT — custom app with OAuth (setup guide)
Claude — Desktop connector and Claude Code CLI (setup guide)
Cursor, Windsurf, VS Code, Cline, Continue, Zed

See MCP Client for step-by-step instructions per client.

Demo video

Self-hosting

The hosted server at https://app.famulor.de/mcp works out of the box. To run your own instance:

Run locally via npm (stdio)

The server is published to npm as famulor-mcp. Run it as a local stdio MCP server — no hosting, no OAuth:

npx -y famulor-mcp          # or: npm i -g famulor-mcp && famulor-mcp

Point a client at it with a stdio config, passing your API key as an env var:

{
  "mcpServers": {
    "famulor": {
      "command": "npx",
      "args": ["-y", "famulor-mcp"],
      "env": { "FAMULOR_API_KEY": "your-api-key" }
    }
  }
}

For most setups the hosted server (https://app.famulor.de/mcp with OAuth sign-in) is simpler — use the npm package when you need a local, offline, or embedded stdio server.

Deploy to Vercel

git clone https://github.com/bekservice/Famulor-MCP.git
cd Famulor-MCP
vercel --prod

Set these environment variables in the Vercel project:

Variable	Required	Value
`MCP_SECRET`	yes	`openssl rand -hex 32` — encrypts OAuth tokens
`MCP_ISSUER`	recommended	Your public URL, e.g. `https://mcp.example.com`

Point your domain at the deployment. See the DEPLOYMENT.md guide in the repository for stdio mode and full details.

Local development

npm install
echo "MCP_SECRET=$(openssl rand -hex 32)" > .env
npm run dev:http
# server on http://localhost:8787

curl http://localhost:8787/health
curl http://localhost:8787/.well-known/oauth-authorization-server

About MCP

Model Context Protocol (MCP) lets AI assistants securely connect to external tools and data. The Famulor MCP server maps platform APIs to MCP tools so clients like ChatGPT and Claude can manage your voice agents on your behalf — after you authenticate with your API key via OAuth.

Resources

License

MIT — see the GitHub repository for details.

Related pages: Use Cases & Prompts, MCP Client, Famulor Skill.

​Famulor MCP Server

Connect a client

Hosted server

GitHub

​Overview

​Hosted endpoints

​What you can do

​Authentication

​Supported clients

​Demo video

​Self-hosting

​Run locally via npm (stdio)

​Deploy to Vercel

​Local development

​About MCP

​Resources

​License

Famulor MCP Server

Overview

Hosted endpoints

What you can do

Authentication

Supported clients

Demo video

Self-hosting

Run locally via npm (stdio)

Deploy to Vercel

Local development

About MCP

Resources

License