Volcengine RTC MCP. Control live streams and monitor network quality from chat.

You're running big broadcasts—the kind that keep people hooked, like what major platforms do with their live feeds. This Volcengine RTC MCP Server gives your agent the direct control over those streams without you ever having to touch a dashboard. It connects your AI client straight to ByteDance's internal Real-Time Communication (RTC) engine, letting your agent handle everything from moderation to recording just based on natural language commands.

Stream Discovery and Monitoring

You can start by figuring out what’s happening across the whole network. Your agent uses get_active_rooms to list every single RTC room currently broadcasting. Once you know which rooms are running, you'll need to check who's in them; get_room_users pulls a full roster of individual participants for any specific stream.

To make sure the feed is clean and functional, you can query real-time quality metrics using get_quality_metrics, pulling detailed performance data about connection issues or dropped frames on any active room.

User Management and Control

If something goes wrong with a user—maybe they're spamming or acting up—your agent has immediate control. It can forcibly remove problematic participants by executing kick_user. For moderation, you don't gotta deal with bad audio or video feeds; your client can send the mute_stream command to block either the audio output or the video output for any user.

When you need to let them talk again, you just run unmute_stream, restoring whatever was previously restricted.

Content Recording and Transcoding

When you gotta keep a record of what's going down, your agent manages it all. To capture the whole show for VOD storage, it starts the process with start_cloud_record and stops it later using stop_cloud_record. If you need to mix content—say, combining three different camera feeds into one stream—the server handles that setup by calling start_transcode, initiating Cloud MCU (Multi-Channel Unit) streaming transcoding.

When the mixing is done or resources are tight, your agent can halt any active process using stop_transcode.

How It Works in Practice

When you hook these tools into your client, your agent translates a plain English request—like "mute John's audio and start recording the room"—into the necessary sequence of API calls. You don't just get data; you control the entire streaming lifecycle: you find the rooms (get_active_rooms), check who's in them (get_room_users), diagnose if the connection is shaky (get_quality_metrics), kick out troublemakers (kick_user), mute their feeds (mute_stream), and finally, capture the whole thing by starting cloud recording or setting up multi-channel mixing.