Zhang Pinpin · Hands-On AI Coach
← Back to home

See a Feature You Like? Have Your Own AI Learn It—Wiring Tencent Meeting into Claude Code

Long-Form Video · EP0059 July 13, 2026 5:57
What this episode covers

You see a genuinely useful feature in someone else's software. What's your first reaction? Most people envy it, or just switch to that tool. But there's another path: have the AI you're already fluent with learn that capability.

This episode walks that path end to end. I'm in Tencent Meeting every day and rely on AI for notes, so I had Claude Code research how to wire Tencent Meeting in, compare four approaches, pick the one that fit me, and then walk me through connecting it. The process doesn't matter. This "have AI steal the trick" habit is the part you can use over and over.

First, something that surprised me recently: I gave Tencent’s WorkBuddy a serious workout. The Hunyuan 3 model is free on it right now, and a lot of everyday office work comes out well as long as you pick the right expert and skill.

But the thing that really caught my eye is its connectors—it packages MCPs for many common apps, especially Tencent’s own, into one-stop integrations. Click, bind, and the MCP is hooked up in one shot. I’m in Tencent Meeting every day and I lean hard on AI for meeting notes, so this saves me real work. I tick the Tencent Meeting connector on a task, then say “a while back I had a meeting with so-and-so, what did we discuss” or “what are our next action items.” It finds the transcript for the right meeting and answers the way I asked. For someone new to AI, that experience is genuinely good.

But my main driver is still Claude Code

The problem is that my daily driver is still Claude Code. I could wire Sonnet or Fable 5 into WorkBuddy, but to be straight with you—I can’t afford it, roughly ten to twenty yuan a minute.

So I started thinking: Tencent Meeting offers its own MCP service, so could I hook it into my Claude Code too?

This is one of the most common mental habits in how I learn and use AI: I have a need, I see another piece of software has already solved it, so I go talk to the agent I use every day and add that capability to my own workflow. The steps themselves don’t matter much in the end. It’s this line of thinking, this way of talking to the thing, that’s worth borrowing.

Letting Claude Code do its own research

First, to clear up a question you may have about my interface: this is neither a terminal nor a desktop client. Claude Code bans have been pretty severe lately, so the safer play is terminal-only, staying away from the client and the web version. But the terminal experience isn’t great, and a lot of the functionality isn’t readable at a glance. So back in late February I wrote my own web version of Claude Code. It’s Claude deployed on a cloud server, and through this page I can see the files on that server, get all my email, and switch between “terminal” and “text chat” however I want, with several windows running in parallel.

Back to the point. Here’s how I briefed it:

I use Tencent Meeting a lot. Right now I paste the transcripts in by hand, or download them and hand them to you for a summary and notes. I want you to build a multi-agent team, search concurrently, and dig into whether Tencent Meeting exposes an MCP you can connect to from outside, so that from here on you can query my recent meeting records and pull the related documents yourself. When the research is done, don’t jump to execution. Give me an operational checklist first and walk me step by step through binding and authorizing the MCP.

The thinking is the same line as before: have AI learn how other people did this, then break it down and walk me through doing it for the first time.

The research came back with several options:

  • Tencent Meeting’s official MCP: personal accounts only, and it asks for sweeping permissions;
  • Tencent’s official open-source CLI (skill): once authorized, AI calls the corresponding functions;
  • Third-party MCPs: built by individual developers, and no longer maintained—ruled out immediately;
  • Plus the underlying Open API, which you can build against yourself.

Some of these only work with personal accounts; some need a business or enterprise plan. I’m on a personal Pro plan, and my need is simple—pull the latest meeting’s transcript or recording. So I had the AI weigh my situation and judge which approach fit me best. After I pressed it, it recommended the second one: authorize the device, then use the official open-source CLI.

It SSHed into my local machine and finished the job itself

My cloud box has a modest spec, so for a lot of tasks I have it connect to my local machine and run things there. Same here: its instructions had me run an installation check locally. But it already has access to that machine, so there was no reason for me to type the commands myself. Just let it keep going.

So I told it: you have authorization on my local machine, so SSH over, run the commands, verify whether it worked, and when you’re done tell me what I can do now. Then we’ll continue.

Sent it, waited a moment. It connected over, ran every command, and everything was ready except one authorization click from me. I flipped the authorization on—that fast? Took another look, and it really was working. It was already inside my Tencent Meeting account, it could see my last several meetings, and it had pulled the transcripts over. It hit some traps along the way, the usual encoding problems among them, but it diagnosed and fixed them itself. At the end it went ahead and wrote up a capability list of what this official CLI can do.

I checked the clock: the whole MCP hookup took about half an hour. So in fairness—WorkBuddy’s one-click integration is still less work, and no wonder, they’re sibling products in the same family.

Once it’s connected, the possibilities open up

But once this MCP is connected, what you can build on top of it opens way up. I floated another request right away:

So in theory, could you poll on a schedule and detect when a new meeting shows up? And when there is one, turn the transcript into the kind of web-page meeting notes I’m used to, pull out the follow-up action items, and drop them in my calendar? Is all of that doable? For now just tell me whether it’s feasible—don’t build it yet.

See, once you’ve plugged in an MCP capability, you can describe a complete, idealized working state and have it build that for you. As for what comes after—turning a set of meeting notes into a consulting report you can hand a client as-is—I covered that in an earlier episode, so I won’t expand on it or sit here waiting for it to finish.

It came back fast: feasible, and most of the functionality it had already verified supports the idea directly.

So in the age of AI, it really is true—the only limit is what you can’t think of, not what can’t be done.

When you see something another piece of software can do, don't sit there envying it and don't rush to switch tools—have the AI you already use learn that capability and wire it into your workflow. The process doesn't matter; the thinking is what's worth borrowing.