RTSP Server Component

The rtsp_server component streams video from an ESP32 Camera over RTSP, the standard protocol used by video players and surveillance software. This allows tools like VLC, ffplay, go2rtc or Frigate to consume the camera feed directly, without any ESPHome-specific integration.

Video is delivered as Motion JPEG (the format produced by the camera hardware) carried over the RTSP connection itself (TCP-interleaved transport). Clients requesting UDP transport are asked to fall back to TCP, which every common player does automatically. The camera streams only while at least one client is playing.

WARNING

The connection is plain TCP with no encryption and no authentication; allowed_ips filters by address but does not authenticate the peer. If it is omitted, any host that can reach the port can watch the camera. Only use this component on a trusted network.

# Example configuration entry, next to a configured esp32_camera
rtsp_server:

Configuration Variables

port (Optional, int): The port the RTSP server listens on. Defaults to 554, the standard RTSP port.
max_connections (Optional, int): The maximum number of clients that can stream at the same time, from 1 to 4. Each connection reserves about 900 bytes of memory and a socket for the whole run, so keep this low on memory-constrained boards. Defaults to 2.
allowed_ips (Optional, list of IPv4 networks): Only peers from these networks may connect; others are refused at accept time. Each entry is a single address (192.168.1.50) or an address/prefix network (192.168.1.0/24). When omitted, every host that can reach the port is accepted.

The component needs an ESP32 Camera that delivers JPEG: the default pixel_format: JPEG, or another colour format converted with jpeg_quality; GRAYSCALE is not supported. The resolution may be at most 2040 pixels per side. The configuration is rejected otherwise. A client whose connection makes no progress for half a second drops its current frame and takes no new one until the connection drains again, so it cannot slow the camera down for the other consumers.

Watching the Stream

The stream is available at rtsp://<device-ip>:<port>/. For example:

ffplay -rtsp_transport tcp rtsp://192.168.1.100:554/

or open the same URL as a network stream in VLC. For the lowest latency with ffplay, disable its internal buffering:

ffplay -fflags nobuffer -flags low_delay -probesize 32 -analyzeduration 0 -rtsp_transport tcp rtsp://192.168.1.100:554/

Throughput

A 640x480 frame is 6 to 19 KB, far more than the 5.7 KB that lwIP queues per TCP connection by default, so the sender has to wait a network round trip every few segments. Over Wi-Fi that limits the stream to about one frame per second and keeps dropping frames. The component therefore requests the high-performance networking settings, which raise the send buffer and the Wi-Fi driver buffers; in the same test they lifted a Wi-Fi client to about 5 fps and an Ethernet client to the camera’s 10 fps (measured on an ESP32-S3 with an OV5640 at 640x480, 10 fps, over Wi-Fi and Ethernet). The larger TCP buffers cost heap while clients are connected; the Wi-Fi driver buffers are reserved for the whole run. Set network: enable_high_performance: false to turn this off on a board that cannot spare the memory, or network: tcp_send_buffer to pick a buffer size by hand.

Integrating into an NVR

The RTSP URL can be used directly as a camera source in surveillance software such as Frigate, go2rtc, Zoneminder or Blue Iris. Configure the source as an RTSP camera with TCP transport and the URL rtsp://<device-ip>:<port>/.

Home Assistant