Kim, this is so incredibly generous and thoughtful 😭 Thank you so much for taking the time to share Amplifiers to your audience. It means the world to me💜 so grateful
Hi Kim! Wonderful post! Question: 🙋🏻♂️ Fornthe Amplifiers MCPs, will the token usage get used up faster? Currently I’m in the Pro plan, and I just created my very first MVPs (2 actually) I maxed out FABLE 5, hehe… but I’m 33% away from maxing out my entire token usage for the week. 🤦🏻♂️ in other words, I may be considering upgrading to the $100/month Max plan.
A lot of that will depend on the Amplifier and the model (obvs, lol).
I’m on the Max plan (worth it… and dammit, now because of Fable I want to upgrade! haha), and I’ve never noticed Amplifiers using a ton of tokens.
This is a great question for @Daria Cupareanu.
However…
I just checked with Claude 😉 (maybe you can verify Daria?)
“Your Claude conversation (the back-and-forth with me) uses your Claude Pro subscription tokens. That part I can't see exact counts for either, but it's covered by your subscription.
The MCP tool calls (like search_pinterest, run_tool) go through Daria's infrastructure. Those costs are on her side, not yours. That's part of what your AIBMM subscription covers.
The image generation (the carousel slides) routes through Daria's MCP to OpenAI's GPT Image 2 API. You do need to put your OpenAI API and Gemini’s Nano Banana API keys in.
So from your perspective as a subscriber, the cost is your Claude Pro plan plus your $12/month AIBMM subscription. You're not paying per-tool-call or per-image (outside of your cost through Google for Nano Banana or OpenAI's API).”
So basically you’re just paying for your conversation with Claude through your subscription.
Amplifiers includes prompts (like Voice DNA), workflows (step-by-step processes that combine prompts + tool calls), and tools like the ones Kim mentioned in the article.
The prompts and workflows can be quite long, which is part of why they produce such good outputs, because they’re carefully designed to guide Claude through the right sequence of steps.
When you use a research tool - for example “pull Reddit discussions about X using amplifiers” - the tool call itself runs through our infrastructure, so it doesn’t use up your Claude subscription. However, once the results come back, Claude still needs to interpret, summarize, organize, analyze them based on what you asked for and that part does use your Claude allowance (but that’s also the part that makes the query useful to you).
The same applies to image generation. I’ve added prompts behind the image-generation tools so that when you ask Claude to create an image with amplifiers (using GPT), it can transform your idea into a stronger image-generation prompt so you get a better result.
So yes, using Amplifiers can use more of your Claude allowance than a very short prompt. But in many cases, it’s still more efficient than going back and forth with Claude until you get a good-quality result.
For example, the Voice DNA skill builder prompt is preeeetty long but you only use it once. After that, you have a skill that helps Claude write much more like you. Without it, you’d have to iterate with Claude every time you sit down to write, trying to get the text closer to how you see it.
Kim I am so grateful you responded! Wow! Lol! 😅. The info you provided was great and definitely helps me. I found you via @Daria Cupareanu Substack and I’m glad I did. Tons of information you guys provide! I appreciate the follow too! Happy Monday!
You and Daria are BOTH blowing my mind…thank you for this really great breakdown. Otherwise I would never have even touched MCPs or had a clue about how they work. The voice piece you want us to start with makes perfect sense, and losing my voice with Ai is a big concern of mine. Thank you, Kim.
Yes!!! You are reading my mind! My favorite tool (Magpie, for saving my own doc context across all the AI chats) is an MCP and there is SO MUCH UNTAPPED POTENTIAL.
I think the way Daria is doing this is early for how all creators should be doing for monetizing their expertise within a world of AI chat. Love this breakdown, thank you!
This got me thinking, Kim. Treating MCPs as digital products feels right, but I think the product surface has to include the risk surface too: what the tool can touch, whether the action can be undone, and what evidence the user gets when it runs. Once the model has a tool, the UX question and the blast-radius question meet. That is the angle I took here: https://petermccannstrain.substack.com/p/tools-give-models-hands
I love that you dug into just what Daria has created here and why it matters. It's such a HUGE library to choose from!! I really need to move away from my usual handful and test out all the other amplifiers available too.
I think the Amplifiers library is probably one of those things I’ll jump into weekly to see what I can try/test relative to what I’m working on (or just play, lol).
I asked Claude for a few good analogies, I liked this one the best 😉
“An API is like a library card that only works for one specific library, and you (a programmer) have to already know exactly which shelf, which book, and which page to ask for. MCP is more like a universal library card that works everywhere, and it comes with a librarian who can go find the right book for you, even if you don't know which library has it.”
I love this! When I asked Claude what was included in Amplifiers, it only returned 10 items, so I’ve had to keep digging to uncover the rest. Daria has put together an incredible catalog for us, packed with more than we could possibly need, but YOU did a phenomenal job listing it all, Kim. Huge thanks for sharing this!
Great write up, but one aspect of MCPs not surfaced that people should be aware of is if any one of those MCP implementations has a vulnerability it greatly increases the odds of an attacker gaining access to data across all connected platforms simultaneously.
If the MCP pulls "profiles, posts, comments, transcripts, ad libraries, and audience demographics." you are trusting the MCP provider (in this case, Daria’s infrastructure) with the ability to scrape and process sensitive data from your connected accounts. There is no mention of where this data is stored or how long it is retained.
My takeaway here is more around awareness and understanding the trade offs that come with the convenience.
One of the best ways to hedge against such vulnerabilities spreading is by deploying an MCP gateway, which acts as an enforcement layer.
This acts as a "semantic firewall" that inspects intent before execution.
Its function is to intercept every tool call, and evaluate the agent's intent against its defined purpose, blocking actions that deviate from the authorized workflow.
Doing a deeper write up on this, so if anyone is interested about balancing security with MCP function, feel free to follow for when its ready.
Kim, this is so incredibly generous and thoughtful 😭 Thank you so much for taking the time to share Amplifiers to your audience. It means the world to me💜 so grateful
You’re so very welcome!
Thanks for all you do & share. And of course, who you are! 🥰
Hi Kim! Wonderful post! Question: 🙋🏻♂️ Fornthe Amplifiers MCPs, will the token usage get used up faster? Currently I’m in the Pro plan, and I just created my very first MVPs (2 actually) I maxed out FABLE 5, hehe… but I’m 33% away from maxing out my entire token usage for the week. 🤦🏻♂️ in other words, I may be considering upgrading to the $100/month Max plan.
Thanks so much,
A lot of that will depend on the Amplifier and the model (obvs, lol).
I’m on the Max plan (worth it… and dammit, now because of Fable I want to upgrade! haha), and I’ve never noticed Amplifiers using a ton of tokens.
This is a great question for @Daria Cupareanu.
However…
I just checked with Claude 😉 (maybe you can verify Daria?)
“Your Claude conversation (the back-and-forth with me) uses your Claude Pro subscription tokens. That part I can't see exact counts for either, but it's covered by your subscription.
The MCP tool calls (like search_pinterest, run_tool) go through Daria's infrastructure. Those costs are on her side, not yours. That's part of what your AIBMM subscription covers.
The image generation (the carousel slides) routes through Daria's MCP to OpenAI's GPT Image 2 API. You do need to put your OpenAI API and Gemini’s Nano Banana API keys in.
So from your perspective as a subscriber, the cost is your Claude Pro plan plus your $12/month AIBMM subscription. You're not paying per-tool-call or per-image (outside of your cost through Google for Nano Banana or OpenAI's API).”
So basically you’re just paying for your conversation with Claude through your subscription.
@Kim Doyal is 100% right about this.
Amplifiers includes prompts (like Voice DNA), workflows (step-by-step processes that combine prompts + tool calls), and tools like the ones Kim mentioned in the article.
The prompts and workflows can be quite long, which is part of why they produce such good outputs, because they’re carefully designed to guide Claude through the right sequence of steps.
When you use a research tool - for example “pull Reddit discussions about X using amplifiers” - the tool call itself runs through our infrastructure, so it doesn’t use up your Claude subscription. However, once the results come back, Claude still needs to interpret, summarize, organize, analyze them based on what you asked for and that part does use your Claude allowance (but that’s also the part that makes the query useful to you).
The same applies to image generation. I’ve added prompts behind the image-generation tools so that when you ask Claude to create an image with amplifiers (using GPT), it can transform your idea into a stronger image-generation prompt so you get a better result.
So yes, using Amplifiers can use more of your Claude allowance than a very short prompt. But in many cases, it’s still more efficient than going back and forth with Claude until you get a good-quality result.
For example, the Voice DNA skill builder prompt is preeeetty long but you only use it once. After that, you have a skill that helps Claude write much more like you. Without it, you’d have to iterate with Claude every time you sit down to write, trying to get the text closer to how you see it.
Thanks for clarifying, Daria 😊
Kim I am so grateful you responded! Wow! Lol! 😅. The info you provided was great and definitely helps me. I found you via @Daria Cupareanu Substack and I’m glad I did. Tons of information you guys provide! I appreciate the follow too! Happy Monday!
Haha, that’s a little meta (you found me through Daria, I love it!).
You’re so welcome. She keeps me inspired!
Happy Monday to you, too!
Oh this is perfect! And makes sense too. I appreciate the response Daria!!
You and Daria are BOTH blowing my mind…thank you for this really great breakdown. Otherwise I would never have even touched MCPs or had a clue about how they work. The voice piece you want us to start with makes perfect sense, and losing my voice with Ai is a big concern of mine. Thank you, Kim.
You’re so welcome, Elizabeth,
I LOVED the Voice DNA exercise from Daria! I just sat and dictated my answers. The pattern recognition, the frameworks… it was so fantastic.
*I’ve also updated my voice skills since then (also just through dictating!). I love the iterative process.
I’ve been wondering about MCPs and what it really does and how it works. Incredible post Kim.
Oooh, these would be perfect for you, Karen!!!
And thank you!
Literally can’t stop thinking about it. So distracting from today’s work!
Haha… well, if you need help girl, I’ve got you!
KIM! Thank you so much. I def will ask!
Yes!!! You are reading my mind! My favorite tool (Magpie, for saving my own doc context across all the AI chats) is an MCP and there is SO MUCH UNTAPPED POTENTIAL.
I think the way Daria is doing this is early for how all creators should be doing for monetizing their expertise within a world of AI chat. Love this breakdown, thank you!
Let’s talk 👀
YES!!!
I’d love to chat. Just DM me and we’ll go from there!
This got me thinking, Kim. Treating MCPs as digital products feels right, but I think the product surface has to include the risk surface too: what the tool can touch, whether the action can be undone, and what evidence the user gets when it runs. Once the model has a tool, the UX question and the blast-radius question meet. That is the angle I took here: https://petermccannstrain.substack.com/p/tools-give-models-hands
Thanks, Peter.
And you’re absolutely right about what the tool can touch. I’m grateful AI makes it easier to look for risks and try to get ahead of them.
I’ll read your article, thanks for sharing.
This is blowing my mind! I need to read the whole thing again. It’s landing at *exactly* the right time for me! ❤️🙏🙌
I’m so glad, Tracy!!!
Holler if you have any other questions. 😊
I love that you dug into just what Daria has created here and why it matters. It's such a HUGE library to choose from!! I really need to move away from my usual handful and test out all the other amplifiers available too.
Same, girl, lol.
I think the Amplifiers library is probably one of those things I’ll jump into weekly to see what I can try/test relative to what I’m working on (or just play, lol).
Okay now you know I have to dig into the ones I am missing action on. And now the wheels are spinning on the other end of the spectrum too.
Hahaha…
Girl, as soon as that came to me (OMG! This is a new digital product type), my wheels started spinning too!
Looking forward to catching up next week and seeing what you’ve cooked up!
Is this like an API with the social media platforms being the libraries?
Kind of…
I asked Claude for a few good analogies, I liked this one the best 😉
“An API is like a library card that only works for one specific library, and you (a programmer) have to already know exactly which shelf, which book, and which page to ask for. MCP is more like a universal library card that works everywhere, and it comes with a librarian who can go find the right book for you, even if you don't know which library has it.”
Hope that helps, David!
Thanks Kim
Amazing that we have the ability to connect all the programs we work with into one fluid system.
It totally is!
And then be able to analyze it all and make sure it’s cohesive (if necessary).
I love this! When I asked Claude what was included in Amplifiers, it only returned 10 items, so I’ve had to keep digging to uncover the rest. Daria has put together an incredible catalog for us, packed with more than we could possibly need, but YOU did a phenomenal job listing it all, Kim. Huge thanks for sharing this!
My pleasure, Anna,
Yeah, logging into your Amplifiers account, then clicking the 'What Amplifiers Can Do’ has the entire list!
https://getamplifiers.com/catalog
I did the same thing from within Claude (asked it to list them) and got a similar response.
Would love to hear which ones you use!
There is a catalog!!!??? Okay, that’s a different story that I (and Claude) completely missed. Will report back 😍🙏🏻
This is where I think a lot of creators are going to have a lightbulb moment.
Your competitive advantage was never the information because info became cheap.
Your advantage is the mental models you use to make decisions.
MCPs might become a way to finally productize that.
Always breaking it down for us like a boss.
Thank you, Kim.
You’re welcome, Neela,
Love the “Always breaking it down for us like a boss.” haha…
Great write up, but one aspect of MCPs not surfaced that people should be aware of is if any one of those MCP implementations has a vulnerability it greatly increases the odds of an attacker gaining access to data across all connected platforms simultaneously.
If the MCP pulls "profiles, posts, comments, transcripts, ad libraries, and audience demographics." you are trusting the MCP provider (in this case, Daria’s infrastructure) with the ability to scrape and process sensitive data from your connected accounts. There is no mention of where this data is stored or how long it is retained.
My takeaway here is more around awareness and understanding the trade offs that come with the convenience.
One of the best ways to hedge against such vulnerabilities spreading is by deploying an MCP gateway, which acts as an enforcement layer.
This acts as a "semantic firewall" that inspects intent before execution.
Its function is to intercept every tool call, and evaluate the agent's intent against its defined purpose, blocking actions that deviate from the authorized workflow.
Doing a deeper write up on this, so if anyone is interested about balancing security with MCP function, feel free to follow for when its ready.
Thank you so much for this, Adam!
Yes, I would LOVE to read your deeper dive when you write it (subscribing!).
Super helpful, thanks so much.
thanks for explaining this so well (and defining what MCP actually means! LOL)! and wow, it sounds AMAZING. and I love your image carousel!
You’re so welcome, Cailtin,
And thank you! You could do some pretty unique MCPs with your work!
This was a perfect Monday morning read to get my brain thinking outside the box. Thank you for explaining it!
My pleasure, Tiffany!
Let me know if you run the prompt, I’d be curious to see what the output is!