pipepie/pipepie

**Self-hosted, encrypted tunnel for webhooks, AI pipelines, and local development.**
**End-to-end Noise encryption, pipeline tracing, web dashboard.**
Why Pipepie?
- **End-to-end encrypted** — Noise NK protocol. The relay server never sees your data.
- **AI-first** — auto-detects Replicate, fal.ai, RunPod, Modal, OpenAI webhooks. Pipeline tracing with timeline visualization.
- **Self-hosted** — your server, your domain, your data. No vendor lock-in.
- **Fast** — 577 req/s parallel, 2ms overhead. Protobuf + zstd + yamux.
- **MCP server built-in** — pie mcp gives Claude, Cursor, and AI tools direct access to tunnels, requests, and pipeline traces.
- **Beautiful CLI** — Dracula theme, interactive forms, 17 commands.
- **Zero-config start** — pie setup on server, pie login on client, done.
Install
Quick Start
Server (your VPS, one time)
Interactive wizard handles DNS, firewall, TLS (Let's Encrypt auto or Cloudflare), nginx detection, systemd service — everything.
Client (your dev machine)
AI Tool Presets
Features
Tunneling
Webhook Inspection
AI Pipeline Tracing
Pipepie auto-detects webhooks from AI providers — no configuration needed:
| Provider | Detection | What's extracted |
|----------|-----------|-----------------|
| Replicate | webhook-id header + payload | Job ID, status, predict_time |
| fal.ai | x-fal-signature header | Request ID, status |
| RunPod | UPPERCASE status in payload | Run ID, execution time |
| Modal | call_id in payload | Call ID, status |
| OpenAI | batch_ prefix in ID | Batch ID, model |
| MCP | JSON-RPC 2.0 method field | Tool name, call ID |
Webhooks from the same pipeline are auto-correlated into traces:
Or use headers for manual correlation:
Multi-Service Config
Multi-Account
Server Management
Architecture
**Noise NK** — server authenticated by public key. Know the key = have access.
**yamux** — multiplexed streams, no head-of-line blocking.
**Protobuf** — binary wire format, ~10x smaller than JSON.
**zstd** — bodies >1KB auto-compressed.
**SSE/streaming** — pass-through without buffering (Vercel AI SDK, Ollama compatible).
MCP Server (Claude, Cursor, AI tools)
Pipepie includes a built-in Model Context Protocol server. Your AI tools can inspect webhooks, replay requests, manage tunnels, and debug pipelines — directly from the chat.
Claude Desktop / Cursor config
Add to claude_desktop_config.json or Cursor MCP settings:
**13 tools available:**
| Tool | What it does |
|------|-------------|
| overview | Dashboard with all tunnels, stats, success rates |
| list_tunnels | All registered tunnels with online/offline status |
| tunnel_status | Check if a specific tunnel is online |
| connect | Start a tunnel (e.g. port 3000 → public URL) |
| disconnect | Stop a running tunnel |
| active_tunnels | List tunnels running in this session |
| list_requests | Recent webhook requests for a tunnel |
| inspect_request | Full request details: headers, body, response |
| replay_request | Re-send a captured webhook |
| pipeline_traces | AI pipeline execution traces |