In this video Michael Elliott warns freelancers that Upwork’s official MCP (Machine Collaboration Platform) is reshaping how proposals are written, submitted, and evaluated. He shares his own recent experiment where his MCP triaged 24 invitations, secured a $1,500/month client, and sent 50 tailored proposals in one week, achieving a 26% response rate (12 conversations). Elliott explains that the MCP connects freelancers’ tools to Claude, automating job search, proposal drafting, and follow‑up, while Upwork’s own AI assistant, Uma, helps clients write job briefs and screen applicants. He argues that most proposals are now read first by AI on both sides, with humans only reviewing summarized shortlists, making volume less important than verified outcomes (reviews, earnings, job success). He outlines practical steps to build and discipline the system, notes platform policy limits (human confirmation required for binding actions), and predicts a future where dual AI agents negotiate job posts. The video ends with a call to watch a full tutorial and subscribe.
Chapters
If you're not using the Upwork MCP right now, you're about to get left behind. Not in a year, not when it's trending— now. Sounds harsh, but last Tuesday my system triaged 24 invitations, held the top proposal slot for 5 jobs, and closed a $1,500 a month client while I was on the school run. And every freelancer that you're bidding against can install the exact same thing this afternoon. And this isn't some third-party hack that Upwork bans next month. This is Upwork's own tool, and it has quietly rewritten the rules of the entire platform. There's a part of this video about who actually reads your proposals now. It's uncomfortable. Every other video on this topic leaves it out, and it might be the only part that matters. I've made half a million on Upwork, and I've done this the manual way for years. Well, that era, it's just ended. Last week I applied for over 50 jobs. 50. Not copy-and-paste spam. 50 tailored proposals that matched the job brief. They weren't copy-and-paste spam. It was 50 tailored proposals, each one matched against my niche, my rate, my win history, drafted in my voice, and sent with my thumb on the button every single time. 26% of them got a response. That's 12 conversations with real clients, with real money. From one week of bidding. You do the maths on the freelancer sending 5 proposals a week by hand. Same skill, same portfolio. The engine behind that is the Upwork MCP. If you've never heard of an MCP, it's just a way of connecting AI to your tools. And Upwork built an official one. You plug it into Claude and everything that you normally dig through the platform for is sitting in one conversation. with an LLM. I mentioned applications, but your connects balance, your open proposals, your close rate, where you rank for your keywords. The morning that I set mine up, it found 1,856 live jobs matching my search. I used to find those by scrolling. And people, they get this part wrong. The tool doesn't win jobs. Your skills, they win the jobs. But what it does is make you systematic. Found faster, drafted sharper, followed up on time every time. But the market doesn't care about how the volume arrives, because when freelancers running this flood every good job within minutes of it being posted, the boosted slots are gone before your lunch break. You're not losing to better freelancers, you're losing to the ones who are plugged in, and you're bidding on their leftovers. That's the half of this tool that everyone's talking about. But the upper half, well, that's the one that should actually worry you because Upwork didn't just build this for us freelancers. I know you've read a job post recently and something about it just felt a little off. Clean structure, tidy bullet points. The ideal candidate will. That wasn't a business owner typing at midnight. That was AI. Upwork's own assistant, Uma, helps clients write their brief, and plenty more clients paste their brief straight out of ChatGPT. And what does that mean? Well, a big slice of your job feed is already machine written. You've seen it. You just haven't said it out loud. Now go one level deeper. On the client side of the MCP, Upwork's own documentation says a hirer can search freelancers, shortlist them, post the job, invite talent, and review proposals, all without leaving AI. Those are 24 invitations that my system triaged last week. I have no idea how many of them a human actually chose to send me. Some of those clients, they were running the exact same class of system that I am, pointed in the other direction. And the hiring side will automate faster than what our side will. Stay with me because I need you to think about why. A freelancer automates to save an hour of scrolling, but a client posting a $200,000 a year problem automates to avoid reading. 80 proposals, 60 of which open with, hey, I hope this finds you well. Their time costs more, so the machine earns its seat much quicker. Which raises the question that most freelancers, we haven't dared to ask out loud. If their AI system wrote the job post and their AI screens the applications, who exactly is reading your proposal? Just quickly, guys, before I answer that, do something for me. Open the last job that you applied for. Read that brief again, but properly this time. Can you tell whether a human or AI wrote it? Comment human or AI with your honest verdict because I want to see how many of you can actually spot it. And if you believe using AI on your applications is cheating, say that down there too. You're about to lose work to people who don't share that principle. And well, I'd rather argue with you now than watch that happen. So who's actually reading your proposal? Increasingly, nobody. Nobody human, not on first pass. So your AI drafts the application and their AI, well, it wrote the brief, screens the shortlist, and then it summarizes all of the candidates together. 2 language models negotiating either side of a job post while the humans read the summary at the end. That's not a prediction that I'm making for 2030. The plumbing for it is live on Upwork right now. I've watched it run from both directions. I've used it from both directions. Upwork's documentation, it asks itself the obvious question, can an AI agent hire on my behalf without me? Their answer is 2 words, not today. Not never, but not today. Today, every binding action still needs a human to confirm it. Every proposal that I send, I approve with my own thumb, and that one rule is what keeps your account alive. But read that sentence again and tell me which direction that this platform is heading in. It wasn't that long ago that Upwork was anti-AI, would ban you for using third-party tools, but now they've got their own official MCP, which has no danger of that. Picture where it lands. A client's agent posts the job at 9 in the morning, screens 40 machine-written applications by quarter past, and puts 3 names in front of the human before their coffee's gone cold. When that arrives, and Not Today tells you that it will, the freelancer's still typing proposals into a text box. They aren't competing at a disadvantage. They're not even in the race. I got this wrong myself at the start. I was skeptical because I know Upwork's history with AI on the platform, and I understand the concerns around it. On day 1, I actually got greedy. You see, I stacked 20 calls from my MCP to Upwork into one run, and I hit their rate limit. 90 seconds of staring at the screen wondering if I'd just torched a $500K profile. Well, I hadn't, but I reined it in, capped everything, and the discipline is now welded into how I run it. Here's what actually matters in a machine-read market. The model on the client side isn't charmed by your opener. It reads the data, your win history, your reviews, your verified earnings, your job success score, everything. The exact things that you can't fake on Upwork, and you can't write your way around it. 2 freelancers run exactly the same system, and the one with real outcomes behind them They win because the machine, it amplifies whatever's actually there. Volume gets you seen, but proof gets you hired. This tool hands you the first one and is completely indifferent about the second one. So the freelancers who get eaten over the next year or so, they aren't the ones without AI, they're the ones with nothing underpinning it. Most freelancers will watch Upwork change underneath their feet and call it unfair. They'll write the angry forum posts, keep scrolling the feed by hand, blame AI slop for flooding feeds. And quietly wonder why the invitations dried up. But you're not most freelancers. You watched the shift while there was still time to move, and you now know what the people taking those top slots do. I've already filmed the full build. It's on screen now. Go and check this out. One paste into Claude, 10 questions, and it constructs my entire system around your profile, your niche, and your rules. Go and watch it before you spend another dollar on connects, because every proposal that you send the old way Is the Kinect spent at a disadvantage? If you've liked this video, hit like, hit subscribe. I'll see you in the next one.
