MCP Setup
MCP lets any MCP-capable AI call VideoFree directly for generation, queries, and project management. It starts with the gateway — no separate MCP service to run.
1Confirm the gateway address
Start VideoFree first. The default port is 8765 and the MCP path is fixed at /mcp.
- Same machine: use
http://127.0.0.1:8765/mcp. - Another machine on the LAN: replace
127.0.0.1with the gateway machine’s real LAN IP, e.g.http://192.168.50.8:8765/mcp. Never use0.0.0.0, and never change the path to/sse.
2Paste the client configuration
Add a server named videofree to your AI client’s MCP configuration and paste the JSON below. Some clients require type to be streamable-http — follow your client’s documentation.
{
"mcpServers": {
"videofree": {
"type": "http",
"url": "http://127.0.0.1:8765/mcp"
}
}
}
Connecting from another machine only requires replacing the address with the gateway machine’s IP:
{
"mcpServers": {
"videofree": {
"type": "http",
"url": "http://192.168.50.8:8765/mcp"
}
}
}
3Add authentication when a token is set
The token is empty by default and no auth header is needed. If you configured a gateway access token in workspace Settings, add headers to the server configuration and replace the example value with your real token:
{ "mcpServers": { "videofree": { "type": "http", "url": "http://127.0.0.1:8765/mcp", "headers": { "X-API-Token": "your-token" } } } }
4Refresh tools and verify the connection
After saving, refresh the MCP tool list in your AI client. Have the AI call get_health, list_capabilities, and list_voices first — normal results mean the connection works.
The client must support HTTP POST with JSON responses. Legacy SSE-only clients cannot use this address directly; do not switch the path to /sse.
If the LAN connection fails, check the gateway bind address, the firewall, and that tokens match. A healthy gateway does not mean ComfyUI is connected — if generation fails, re-check the ComfyUI address in Settings.
5Need a smaller tool surface?
The full MCP exposes 30+ tools covering the entire workspace. For fully automated pipelines, you can wrap a narrow MCP containing just the few tools you need over the HTTP API — see Narrow MCP.