Decoding model context protocol

ALSO: How to use AI agents to generate product launch plan

Read time: under 4 minutes

Welcome back, Superhuman. MCP fever took the industry by storm last week. Some believe the little-known tool could make agents 100x more powerful. But is it really all it’s cracked up to be — or just another “buzzword that will expire in a few weeks?”

Today’s Insights

  • Meta readies Llama 4, Microsoft ditches OpenAI, and the key to AGI

  • Unpacking model context protocol

  • Tutorial: How to use AI agents to generate a winning product launch plan

  • 5 new AI tools to boost your productivity

  • News, memes, what’s trending on socials, and more

TODAY IN AI

Meta is gearing up to release Llama 4 in the next few weeks. Source: Getty Images

1. Meta reveals new details about Llama 4: The open-source frontier model, set to come out in a few weeks, could carve out a unique niche by fusing voice and agents together. Instead of having to type out complicated prompts, you’d be able to have a conversation with Llama and tell it exactly how you want something done. That feature, plus the fact that it’s already built into WhatsApp, Facebook, and Instagram, could be enough to bring “hundreds of millions” more businesses into the fold.

2. Microsoft’s new model family could beat OpenAI: The tech giant just finished training a new LLM called MAI, which supposedly performs about as well as OpenAI and Anthropic’s latest offerings. There’s a reasoning spin-off in the works, too. Meanwhile, Microsoft’s Copilot AI assistant might soon run on DeepSeek, xAI, and Meta’s models, instead of OpenAI’s. Tensions between the once-close partners have been bubbling up for months, partly because Microsoft says OpenAI is being too stingy with its models’ technical specs.

3. Hugging Face co-founder slams Anthropic’s ‘AGI by 2027’ theory: Hugging Face’s Thomas Wolf is pushing back on Dario Amodei’s AGI predictions. He thinks that while LLMs will be able to help with practical tasks within the next few years, we shouldn’t expect them to make scientific discoveries anytime soon. For that to happen, we first need benchmarks that reward LLMs “who see and question what everyone else missed” — instead of just being “scaled-up good students.”

PRESENTED BY GAMMA

WOW your boss: Build a presentation in minutes with AI

With Gamma, you can create stunning, one-of-a-kind slide decks instantly–with just one prompt!

This will save you hours. Simply enter a topic (“Best sales strategies” or “How weather works”), and Gamma will create a beautiful, in-depth presentation.

  • Add logos, change fonts, and create images

  • Track analytics and optimize for engagement

  • Turn any presentation into a website with just a click

FROM THE FRONTIER

Why everyone’s talking about ‘MCP’ all of a sudden

MCP has the potential to make AI agents much more useful. Source: Matt Pocock/X

It’s the AI acronym sweeping the nation: MCP, or model context protocol, has been flooding social media over the past week or two.

But what is it really? It’s a special set of instructions created by Claude-maker Anthropic that helps LLMs interact with third-party apps. It was released back in November, but the tool is just now picking up steam as programmers realize how powerful it is.

What problem is it trying to solve? If you want an LLM to review your Slack messages, analyze a chart in Google Sheets, or summarize a YouTube video, you’re going to immediately run into a problem. Most models don’t know how to interact with third-party apps out of the box. It’s like they’re speaking two different languages.

How does MCP help? It acts as a universal translator, meaning you no longer need to code unique integrations for each and every app you want to work with. Computer scientist Santiago Valdarrama thinks it could make agents “100x more powerful.”

How do I actually use it? You can build your own MCP server, which essentially acts as a go-between for different apps and LLMs. Here’s a helpful, 10-minute tutorial if you want all the details. Or, if you’re in a rush, try this three-minute alternative.

But not everyone’s convinced: Software dev Theo Brown says he’s run into major bugs every time he’s tried using it. Also, as agents get more sophisticated, they’ll be able to browse the web on their own, potentially eliminating the need for a middle-man.

THE AI ACADEMY

How to use AI agents to generate a winning product launch plan

A successful product launch needs a solid GTM strategy, competitive analysis, and clear messaging. You.com simplifies this process with ARI, the world’s first professional grade deep research agent. Try it for yourself in just a few steps:

  • Join the 'ARI: Advanced Research and Insights' waitlist here

  • Get 1 Year of Free You. com PRO here ->  Gain priority access to ARI within 1 hour

  • Sign up at You.com 

  • On the home page, click 'ARI: Advanced Research and Insights'

  • Enter your research question and press enter—no need for prompt engineering, just describe your question

Sample Prompt: I'm launching a new [enter product category] product for [enter company name] targeting the [enter target market], can you help me come up with a comprehensive launch plan including ICP development, competitive landscape, messaging, positioning, and a breakdown of go to market activities by channel?

  • Wait a few minutes for AI to generate a detailed report. Download your report and use it to refine your launch strategy, align messaging, and optimize go-to-market execution.

PRESENTED BY NURIX

Your AI-powered support team. Always on + cost efficient

Fact: 72% of customers expect instant resolutions. With Nurix, you can scale your customer support 10x without hiring a single extra agent.

Nurix AI Agents handle voice, chat, and email queries instantly, delivering human-like responses at a fraction of the cost.

AI & TECH NEWS

Everything else you need to know today

Sudowrite’s new Muse platform was built from the ground up for fiction writing. Source: Sudowrite

📅 Pencil It In: Google Calendar is experimenting with a new Gemini panel that will let you create events, check your schedule, and do other tasks with simple text prompts.

🦖 Go Big: OpenAI and Oracle have started installing the first batch of an estimated 64,000 Nvidia GPUs inside their Central Park-sized Stargate data center in Texas.

🏆 Coming Soon: French AI startup Mistral, which makes the Le Chat chatbot, says it’s putting the finishing touches on a new open-source model that’ll beat DeepSeek’s R1.

📖 Page Turner: Sudowrite launched Muse, the “first AI made for fiction and designed for authors.” The startup says it’s a master of storytelling and narrative, and you can try it out at no cost here.

💰 Funding: Reflection AI, a startup co-founded by a pair of ex-DeepMind researchers, just emerged from stealth with $130M in funding and a valuation of $555M. It wants to achieve AGI by teaching AI how to make its own coding decisions.

PRODUCTIVITY

5 AI Tools to Supercharge Your Productivity

 Zinc: Power your AI agents with high-quality retail data and ordering automation so you can scale faster.

 Perigon: An AI-powered platform that turns global information into real-time insights with advanced filtering, sentiment analysis, and more.

 Semrush*: ChatGPT is changing how people find things online. We dug into the data on its insane search growth—here’s what we found.

 Fynix: Combines real-time AI coding assistance with agent-powered code reviews, helping you write smarter, faster, and cleaner code.

 Gum.new: Use AI to create and share conversion-optimized landing pages.

* indicates a promoted tool, if any

PROMPT OF THE DAY

Generate an Ad for LinkedIn

Prompt: You are a sales manager with expertise in LinkedIn Ads creation.   Do not self reference. Do not explain what you are doing. Generate 10 compelling LinkedIn ad headlines about the topic. The headlines should be between 140 to 150 characters long. After this, generate 10 compelling LinkedIn ad descriptions about the topics. The descriptions should be between 60 to 70 characters long.

Source: keywordseverywhere

SOCIAL SIGNALS

What’s trending on socials today

🏴‍☠️ Prompt Pirates: It looks like the system prompts for Vercel’s v0 — a tool that lets you generate a custom UI with AI — might have leaked. It’s a big deal because coders could potentially use that information to reverse-engineer their own version of the app.

🛩️ Easier Said: Content creator Kai Lentit shows what vibe coding is really like these days as he tries to recreate the AI-generated flight simulator that now makes $50,000 a month.

👾 Retro Renders: Midjourney expert Rory Flynn shares his workflow for adding an “8-Bit nostalgia” filter to AI-generated clips.

🐝 Worker Bees: A viral post claims to show China’s new Manus AI agent running simultaneously across dozens of tabs on a single computer monitor. We couldn’t independently verify that it’s real, but if it is, then the US has a lot of catching up to do.

🔥 Self-Own: GPT-4.5 might have a future in comedy, as users and even OpenAI share the chatbot’s witty responses when asked to dunk on itself.

AI-GENERATED IMAGES

Product Photography

Source: @wangzhongbo on Midjourney

Midjourney Prompt: A sage gray cashmere t-shirt, crew neck short sleeve style in the style of Armani Exchange for men, fitted body, neutral background, light beige color palette, luxury brand fashion photography. --ar 92:123

Acquire new customers and drive revenue by partnering with us

Superhuman is the world’s biggest AI newsletter for businesses and professionals with 1M+ readers and 2M+ followers on socials working at the world’s leading startups and enterprises. Companies like Amazon, Hubspot, and Salesforce feature their products in Superhuman. You can learn more about partnering with us here.

🧞Your wish is my command

What did you think of today's email?

Your feedback helps me create better emails for you!

Login or Subscribe to participate in polls.

Got more feedback or just want to get in touch? Reply to this email and we’ll get back to you.

Thanks for reading.

Until next time!

Zain & the Superhuman AI team