Clients

The three first-party PrivacyFlow clients — MCP server for LLM tools, Agent Zero channel for autonomous replies, and the n8n node for visual workflows.

PrivacyFlow ships three first-party clients. They all talk to the same public API and authenticate with the same API key model; the differences are in how you install them and where the work happens.

The three clients

How to choose

• You want an AI agent to read your inbox and reply. Use Agent Zero. It runs a 3-second poller, forwards inbound messages to an Agent Zero context, and ships the agent’s reply back through PrivacyFlow’s send endpoint with per-messenger message splitting.
• You want a chat-style LLM client (Claude, Cursor) to call PrivacyFlow as a tool. Use the MCP server. It exposes four tools (health, verify, poll, send) plus two prompts (reply-to-message, broadcast) and two resources (API docs, contact formats). The LLM decides when to poll and what to send.
• You want to build automation with no code. Use n8n. The Send node sends messages; the Trigger node polls and emits each message as an n8n item you can route through any other node.

Shared conventions

All three clients:

• Authenticate with a single PrivacyFlow API key in the Authorization: Bearer <key> header by default (the MCP server switches to X-API-Key if you set PRIVACYFLOW_API_KEY_HEADER=x-api-key).
• Default to the public API at https://privacyflow.app; override the base URL per-client if you run a self-hosted instance.
• Treat polling as a destructive read — once a message is polled, it’s gone. They all persist before processing.
• Preserve the messenger, contactId, and groupId from polled messages when sending replies. The n8n trigger node surfaces this caveat in its UI; the others do it in code.