Tencent TRTC MCP. Control live video sessions and recordings from your agent.

You're hooking your AI agent up to a massive real-time video comms engine from Tencent TRTC. This isn't just for basic calls; you use it when y’all need automated control over live streams, detailed metrics on call health, and professional archiving capabilities. It lets your agent manage everything from who's in the room to combining multiple feeds into one single output.

Monitoring Call Performance and Room Status

When you run into a flaky connection or just need an audit trail, you can assess the session metrics using describe_call_detail_info. This tool pulls detailed data points about the live video call's quality and performance—you'll get specifics on things like frame drops and bandwidth usage. For a quick status check, you use describe_room_info to pull current details for any specific TRTC room session, figuring out if it's active or when it was even created.

If your agent needs to know who’s actually on the line, you call describe_user_info, passing in a known call session ID and getting back a list of all unique participant IDs. If you need a high-level overview for billing or resource planning, you pull aggregated service data by invoking describe_trtc_usage to see overall usage patterns and volume statistics.

Controlling User Presence in Rooms

Your agent has full control over who's talking shop. If someone needs to be kicked out—whether they’re spamming or the meeting is done—you can force termination of an entire session using dismiss_room, which immediately disconnects every user in that TRTC room. You don't have to dismiss everyone, though. To target specific people, you use remove_user by providing a user’s ID and the active room context.

If you know the string-based room ID but not individual user IDs, you can manage multiple exits at once calling remove_user_by_str_room_id. These tools let your agent surgically control participant lists in real time.

Managing Video Streams and Archiving Content

If y’all need proof of what happened, or if the content is valuable enough to keep, you manage stream archiving. To start recording every video output and activity happening in an active room, you run start_cloud_recording. When the session wraps up, you use stop_cloud_recording to halt that specific cloud task.

For advanced mixing—say, when three different people are talking but you only want one clean feed for a presentation—you initiate the process with start_mcu_mix, which starts multi-channel unit (MCU) mixing and combines multiple video feeds into a single stream source. When the mix is done, you stop it by calling stop_mcu_mix, freeing up those resources.

How It Works Under the Hood

To get started, your agent needs credentials: Tencent SecretId, SecretKey, and SdkAppId. You're meant to inject these into your LLM logic so it treats this server like an automated stream backend. Once connected, your agent can invoke specific tools—like remove_user or start_cloud_recording—to execute complex video operations you otherwise couldn't touch.