webrtc-mcp-server
Peer-to-peer WebRTC communication for AI agents, connecting autonomous coding agents over low-latency WebRTC DataChannels with room-based signaling, video stream bridging, and multi-agent coordination — all exposed as MCP tools.
README
WebRTC MCP Server
Peer-to-peer WebRTC communication for AI agents. Connects autonomous coding agents over low-latency WebRTC DataChannels with room-based signaling, video stream bridging, and multi-agent coordination — all exposed as MCP tools.
TL;DR
npm install && npm run build
# As MCP server (stdio)
node dist/index.js
# As WebSocket signaling server (for external peers)
WEBRTC_SIGNALING_MODE=ws WEBRTC_WS_PORT=8765 node dist/index.js
| Feature | Detail |
|---|---|
| Protocol | MCP v2025-03-26 (over stdio), WebSocket signaling |
| Tools | 16 MCP tools — peers, rooms, streams, signaling, health |
| Concurrency | Worker Thread pool (default 8, max 16) |
| Latency | 0.82ms avg signaling RTT (measured) |
| Sources | RTSP, HLS, RTMP, WebRTC (auto-detect) |
| Tests | 14/14 passing |
What It Does
Multi-Agent Communication
Connect AI agents (Claude Code, OpenCode, Cursor, and any MCP client) over WebRTC DataChannels. Each agent becomes a peer that can:
- Join named rooms for group communication
- Send structured messages (JSON) with ACK
- Broadcast to all peers in a room
- Relay SDP/ICE for WebRTC handshake
Video Streams
Bridge RTSP/HLS/RTMP video streams to WebRTC for real-time frame access:
webrtc_connect_stream(url)→ creates RTCPeerConnection + offer SDPwebrtc_frame_get(stream_id)→ returns latest frame as base64 JPEGwebrtc_stream_status(stream_id)→ health, FPS, throughput metrics- FFmpeg-backed decoding with ring-buffer frame cache
Room-Based Signaling
WebSocket server (ws://host:port) for external peers:
join/leave/list_peers— room membershipsignal— SDP/ICE relay between peersbroadcast— fan-out messages to all room membersping/pong— health check
Quick Start
As a standalone server
# stdio mode (MCP transport)
WEBRTC_SIGNALING_MODE=stdio node dist/index.js
# WebSocket signaling mode (for external peers)
WEBRTC_SIGNALING_MODE=ws WEBRTC_WS_PORT=8765 node dist/index.js
# Both modes simultaneously
WEBRTC_SIGNALING_MODE=both node dist/index.js
As an MCP server (Claude Desktop / Cursor / any MCP client)
{
"mcpServers": {
"webrtc": {
"command": "node",
"args": ["/path/to/dist/index.js"],
"env": {
"WEBRTC_SIGNALING_MODE": "stdio"
}
}
}
}
External peer via WebSocket
// Node.js client
const ws = new WebSocket('ws://127.0.0.1:8765');
ws.on('open', () => {
ws.send(JSON.stringify({
type: 'join',
peerId: 'peer-a',
room: 'my-room'
}));
});
MCP Tools
Peer Communication (6 tools)
| Tool | Description |
|---|---|
webrtc_connect |
Create RTCPeerConnection + DataChannel with a peer |
webrtc_disconnect |
Close connection and release resources |
webrtc_send |
Send structured message via DataChannel |
webrtc_broadcast |
Broadcast to all peers in a room |
webrtc_list_peers |
List all connected peers |
webrtc_peer_status |
Detailed status of a specific peer |
Rooms (4 tools)
| Tool | Description |
|---|---|
webrtc_create_room |
Create a new signaling room |
webrtc_join_room |
Join a peer to a room |
webrtc_leave_room |
Leave a room |
webrtc_signal_relay |
Relay SDP/ICE candidates between peers |
Video Streams (5 tools)
| Tool | Description |
|---|---|
webrtc_connect_stream |
Connect RTSP/HLS/RTMP source → WebRTC offer |
webrtc_frame_get |
Get latest frame as base64 (cached, no re-encode) |
webrtc_list_streams |
List all active streams with health metrics |
webrtc_stream_status |
Detailed streaming metrics (FPS, throughput) |
webrtc_disconnect_stream |
Close a video stream |
Health (1 tool)
| Tool | Description |
|---|---|
webrtc_health |
Overall server health (peers, rooms, workers, uptime) |
Configuration
All config via environment variables or config.yaml:
| Variable | Default | Description |
|---|---|---|
WEBRTC_SIGNALING_MODE |
stdio |
stdio | ws | both |
WEBRTC_MAX_WORKERS |
8 | Max concurrent worker threads (≤16) |
WEBRTC_FRAME_CACHE_SIZE |
5 | Frames cached per stream (ring buffer) |
WEBRTC_MAX_FRAME_BYTES |
500000 | Max JPEG payload (~480KB @ 1920×1080) |
WEBRTC_CONNECTION_TIMEOUT |
30 | Connection timeout in seconds |
WEBRTC_WS_PORT |
8765 | WebSocket signaling port |
WEBRTC_WS_HOST |
127.0.0.1 | WebSocket bind address |
WEBRTC_LOG_LEVEL |
warn | debug | info | warn | error |
WEBRTC_STUN_URL |
stun:stun.l.google.com:19302 | STUN server |
WEBRTC_ALLOWED_URLS |
(auto) | Comma-separated URL allowlist |
See config.yaml for the full default configuration with TURN, rate limiting, and ICE restart settings.
Multi-Agent Workflow
1. Agent A: webrtc_create_room("team-sync")
2. Agent B: {type:"join", peerId:"agent-b", room:"team-sync"} ← WebSocket
3. Agent A: webrtc_connect(peerId="agent-b") → RTCPeerConnection + DataChannel
4. A → B: webrtc_send(peerId="agent-b", data={"task":"review","file":"src/index.ts"})
5. B → A: webrtc_send(peerId="agent-a", data={"result":"✅ no issues"})
6. Broadcast: webrtc_broadcast(data={"type":"status","msg":"deploying"})
Video Stream Workflow
1. webrtc_connect_stream(url="rtsp://camera.local:554/stream1")
→ {stream_id: "cam-123", offer_sdp: "...", ice_servers: [...]}
2. webrtc_frame_get(stream_id="cam-123")
→ {frame: "base64...", timestamp: 1753785600000, resolution: {width:1920, height:1080}}
3. vision_analyze(image="data:image/jpeg;base64,...", question="¿Hay personas?")
→ "Sí, 2 personas detectadas"
Frames are cached in a thread-safe ring buffer — repeated frame_get calls return the same buffer without re-encoding.
Architecture
┌─────────────────────────────────────────────────────────┐
│ MCP CLIENT (Claude Desktop, Cursor, any MCP host) │
│ MCP stdio transport: node dist/index.js │
├─────────────────────────────────────────────────────────┤
│ MCP Protocol Handler (v2025-03-26) │
│ → tools/list, tools/call → dispatch │
├─────────────────────────────────────────────────────────┤
│ WebSocket Signaling Server (ws://127.0.0.1:8765) │
│ → join/leave/list_peers/ping/broadcast/signal │
├─────────────────────────────────────────────────────────┤
│ Worker Thread Pool (8 concurrent, round-robin) │
│ ├─ Worker 1: RTCPeerConnection + DataChannel │
│ ├─ Worker 2: RTSP/FFmpeg → WebRTC bridge │
│ └─ Worker N: isolated per peer/stream │
├─────────────────────────────────────────────────────────┤
│ FFmpeg Bridge │
│ RTSP/HLS/RTMP → raw frames → JPEG (via sharp) │
│ FrameCache: thread-safe ring buffer (5 frames) │
└─────────────────────────────────────────────────────────┘
Development
npm install # install dependencies
npm run build # TypeScript → dist/
npm run dev # watch mode (tsx)
npm test # 14 tests (vitest)
npm run typecheck # tsc --noEmit
npm run lint # eslint
Tests
Test Files 2 passed (2)
Tests 14 passed (14)
| File | Tests |
|---|---|
test/room.test.ts |
9 tests — room join/leave, peer management, broadcast |
test/signaling.test.ts |
5 tests — SDP/ICE routing, health checks |
License
MIT — see LICENSE.
WebRTC is the industry standard for real-time P2P communication (used by Zoom, Google Meet, Discord). This server brings that capability to the MCP ecosystem for multi-agent collaboration.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.