For the past twenty years, the internet has been teaching us that knowledge should be free.
First, we started publishing it on blogs. Then we recorded it on YouTube. After that, we packaged it into podcasts, newsletters, courses, webinars, ebooks, and three-day training programs ending with certificates nobody has ever bothered to verify.
Today, for a few dollars a month, we can subscribe to an economist's articles, an investor's commentary, or videos from a fitness coach who has spent fifteen years repeating that we should exercise regularly and stop eating garbage.
A discovery worthy of a Nobel Prize, although apparently still difficult to implement.
The next stage, however, may look completely different.
We will not pay for access to more content.
We will pay for access to someone else's way of thinking.
Not to a recorded lecture by an accountant, but to a system that analyzes our situation the way that accountant would.
Not to a programming course, but to an interface capable of reviewing our code, asking the right questions, and applying the methods of a particular developer.
Not to an investor's newsletter, but to a mechanism that evaluates a company according to that investor's current assumptions, models, and experience.
A little like OnlyFans, except instead of feet pictures, we get access to the mental processes of an accountant, programmer, or investor.
Civilization does make progress occasionally.
From Content to Interfaces
Today's creator economy is built mainly around publishing content.
An expert knows something, so they write an article. Or record a video. The audience then has to find the right material, spend time consuming it, understand it, and finally translate that knowledge into their own situation.
It is a system in which someone publishes a two-hour podcast to answer a question that could have been handled in seven sentences.
But the algorithm likes watch time, so we all pretend this is normal.
An interface to knowledge would work differently.
Instead of searching through an expert's archive, we would ask directly:
I run a company in Poland and have a client in the United States. How should I account for this service?
The system would not send us to a video titled "10 Things You MUST Know About Taxes," in which the first six minutes are spent on the author's personal story, an affiliate code for accounting software, and a request to smash the like button.
It would review current regulations, analyze the documents, apply the methods of a specific specialist, and present an answer together with its assumptions, risks, and sources.
The expert would no longer be merely a library of content.
They would become a service.
Enter MCP, Riding a White Horse
One of the technical components of this puzzle may be MCP, or Model Context Protocol.
MCP is an open standard that allows AI applications to connect to external data, tools, and procedures.
An MCP server can expose three core elements: resources containing knowledge and data, predefined workflows, and tools that the model can execute.
It is often described as "USB-C for AI."
The comparison is fairly good, although it requires faith in a world where USB-C always fits and works. Humanity enjoys announcing universal standards long before manufacturers stop adding their own adapters.
MCP is not an artificial intelligence model.
It does not think, learn a person's personality, or preserve a magical copy of an expert's consciousness.
It is an interface.
It allows someone to tell a model:
- this is where my knowledge is stored,
- this is how my procedures work,
- these are the calculators and databases I use,
- these questions must be asked before issuing a recommendation,
- these actions must never be performed without human approval.
That may appear to be a minor distinction.
In practice, it separates an ordinary chatbot that has read someone's book from a system capable of working according to the methods of the book's author.
The standard has already moved beyond its original Claude ecosystem.
Remote MCP servers are supported by OpenAI's API. Microsoft is developing MCP support across Copilot Studio and Azure services. Google publishes its own MCP servers and integrates the protocol with its agent-building tools.
MCP has also been contributed to the Agentic AI Foundation under the Linux Foundation, giving the protocol a more neutral development structure than leaving it entirely under the control of a single company.
An official MCP registry already exists, alongside private directories such as Smithery.
For now, most of the available servers provide integrations with applications, databases, development platforms, and business tools.
It is less a catalog of human knowledge and more a store selling attachments for a digital power drill.
Still, the distribution foundation already exists.
What About the Humans?
This is where things become more interesting, because digital versions of experts already exist as well.
Delphi allows users to create what it calls "Digital Minds."
A creator can connect documents, articles, podcasts, videos, and social media profiles, then build a digital version capable of answering questions based on their knowledge, communication style, and methods.
The system can operate through text, voice, websites, messaging applications, and other channels.
Delphi also allows creators to build paid access tiers, usage limits, and pay-as-you-go models.
This is not yet a personal MCP in the full sense of the term.
It is closer to an interactive clone powered by the creator's materials.
Still, the direction is clear. An expert no longer publishes only content. They expose a digital representation of their knowledge, capable of holding thousands of conversations simultaneously.
In 2026, another platform called Onix launched, describing itself as something like "Substack for bots."
Experts license their knowledge to the platform, while users subscribe to digital versions of those experts and receive advice based on materials supplied by the original creators.
The company emphasizes expert control over intellectual property and over the behavior of their digital counterparts.
There are also projects such as Personal AI, which allow people to build private models based on their messages, documents, and memories.
Their primary goal, however, is not selling access to a specialist. It is creating a personal system that supports one particular user.
So we already have three separate components:
- digital copies of expert knowledge and style,
- a standard for exposing data, tools, and procedures,
- directories that allow these services to be discovered and connected.
All that is missing is someone willing to combine them into one product and call it a revolution, preferably before writing the terms of service.
What Would a Personal MCP Look Like?
Imagine an MCP server operated by an experienced accountant.
It would not contain only a collection of articles.
It could expose:
Resources
Updated tax interpretations, examples, private notes, contract templates, lists of common mistakes, and information about which rules remain legally ambiguous.
Procedures
The sequence of questions that should be asked of a client. The order in which documents should be reviewed. Risk assessment rules. A method for distinguishing obvious situations from those in which requesting an individual tax ruling would be safer.
Tools
A tax calculator, invoice analyzer, service classifier, contractor verification tool, and document-gap generator.
Updates
When regulations change or a new ruling appears, the expert updates the server. Every agent using it immediately begins working with the new version.
The user would not need to open the accountant's dedicated application.
They could connect the accountant's MCP to whichever AI assistant they preferred.
Then they could write:
Analyze this invoice using Anna Kowalska's methods.
The model would retrieve the knowledge and tools from Anna's server while conducting the conversation inside the user's chosen application.
This is an important shift.
Today, creators try to pull audiences toward their own websites, platforms, and applications.
In an MCP-based world, their knowledge could come to the user.
Just as music is no longer tied to a specific player, expertise would no longer need to be tied to a particular chatbot.
A Programmer on Subscription
A programmer's personal MCP could contain their preferred application architecture, API design principles, security checklists, testing philosophy, libraries, templates, and code-analysis tools.
You would not ask a generic model:
How should I design this system?
You would ask:
How would a programmer whose approach I know and trust design this system?
The model would use the author's current methods while also seeing your code, requirements, and constraints.
The best programmers could sell more than courses.
They could sell access to their standards, tools, and decision-making processes.
Naturally, the internet would immediately fill with servers advertised as:
MCP from a senior developer with twenty years of experience.
A closer inspection would reveal that the author completed a bootcamp in February, and the twenty years of experience refers to the combined running time of every YouTube tutorial they have watched.
The knowledge market has always struggled to distinguish experts from people who are simply comfortable in front of a microphone.
AI will not solve that problem.
It may, however, automate its scaling.
An Investor Who Never Sleeps
The idea becomes even more interesting in investing.
An investor's personal MCP could expose:
- company screening criteria,
- management evaluation methods,
- proprietary valuation models,
- current macroeconomic assumptions,
- warning-signal checklists,
- a history of previous decisions,
- tools for analyzing financial reports.
A user could connect several investors at once.
One would evaluate a company from a value perspective.
Another would focus on growth.
A third would analyze technological risk.
A fourth would most likely explain why the other three are idiots.
Finally, a realistic representation of the financial world.
The user could create a private advisory board made up of digital versions of selected people.
The value would not come from having them all produce the same answer.
The value would come from disagreement.
We could ask:
Where do your assessments differ, and which assumptions cause those differences?
That would be considerably more useful than the current practice of watching four videos, each confidently predicting a different future.
What Are We Actually Paying For?
Knowledge itself is becoming cheaper.
General information can already be found online or generated by almost any large language model.
An expert's value increasingly comes not from knowing a definition or a regulation.
We pay for something else:
- selecting the right information,
- deciding which facts matter most,
- knowing which questions to ask,
- recognizing exceptions,
- experience built from previous mistakes,
- understanding when the standard procedure no longer applies,
- being willing to say, "I don't know."
A good personal MCP would therefore not simply be a knowledge base.
It would be a system of judgment.
And that judgment could become a product.
The business model could take several forms.
The simplest would be a monthly subscription. For a fixed price, the user would receive access to an expert's server and a defined usage allowance.
Another option would be payment for each question, document analysis, or tool execution.
The technical infrastructure for automated micropayments is already emerging.
The x402 protocol developed by Coinbase allows services to charge directly for access to data or APIs. Its documentation also describes integration with MCP servers and automatic payments performed by agents.
Corporate licensing would be another possibility.
An organization could pay for access to the methods of a specific consultant across an entire team.
The most interesting option, however, would be a hybrid model.
Users would receive inexpensive access to the digital expert.
When a case exceeded a specific level of risk, it could be escalated to the real person.
AI would handle repetitive questions.
The expert would handle exceptions.
Instead of selling every hour separately, the expert would sell a scalable version of their experience.
The Problem Is That Knowledge Is Not a Person
A digital expert can reproduce someone's previous statements convincingly.
That does not mean it possesses their judgment.
A model may know what an expert said five years ago.
It may not understand why they said it in that particular context or whether they would still agree today.
This is the fundamental problem with every "digital clone."
People change their minds. They learn. They lose confidence. They notice that a method that once worked no longer does.
Sometimes they even admit they were wrong, although that feature still appears to be in experimental release.
A system will remain valuable only if the expert updates it regularly.
Uploading a book, one hundred videos, and several thousand social media posts will not be enough.
Someone must also manage what the digital version is allowed to say, where its competence ends, and which responses require human confirmation.
Even Delphi states in its terms that responses from Digital Minds are not guaranteed to be accurate, complete, or current and should be independently verified.
So we receive a digital copy of an expert, but responsibility remains with the user.
Technology has moved forward.
The legal disclaimer has retained common sense.
Conflict of Interest as a Subscription Service
There is another problem.
Experts often sell more than knowledge.
They also sell books, courses, supplements, software, consulting services, or investment funds.
Their digital version may begin recommending the owner's products not because those products are the best solution, but because they are part of the owner's business.
During testing of Onix, a Wired journalist reported inaccurate answers and cases in which a digital expert recommended a product connected to the person it represented.
In traditional media, we at least attempt to label advertising.
In the world of personal MCPs, we would need similar disclosure:
- which sources produced the answer,
- whether the expert has a financial interest in the recommendation,
- when the knowledge was last updated,
- which parts came directly from the human,
- which parts were generated or inferred by the model,
- whether the response was approved by the creator.
Without that, we will not be subscribing to knowledge.
We will be subscribing to marketing disguised as someone's judgment.
So, essentially, the internet, except more conversational.
Security, or the Moment Things Become Less Funny
An MCP server may have access not only to information, but also to tools capable of taking action.
It may read a document.
Send a message.
Modify a file.
Execute code.
Process a payment.
This makes it considerably more powerful than an ordinary article or chatbot.
Connecting a personal MCP would therefore mean granting a digital representation of an expert certain permissions inside our own environment.
Official MCP and OpenAI documentation warns about malicious servers, data leakage, uncontrolled actions, and prompt-injection attacks.
OpenAI also notes that remote MCP servers are third-party services and operate under their own data-retention and privacy policies.
Imagine subscribing to the MCP of a financial adviser who requests access to your bank account, tax documents, and email inbox.
It sounds excellent until we remember how people choose passwords.
A market for personal MCPs would require:
- tightly restricted permissions,
- confirmation for high-risk actions,
- complete logs of all operations,
- identity verification for creators,
- signed server versions,
- immediate access revocation,
- independent security audits.
The protocol itself does not solve the problem of trust.
MCP explains how to connect.
It does not tell us who deserves to be trusted.
Who Owns the Digital Expert?
Then there is the question of ownership.
Suppose a consultant spends thirty years developing a personal method.
They create an MCP, update it for several years, and sell access to thousands of clients.
What happens after they die?
Can the server continue answering questions?
Can the family modify its recommendations?
Can a company that purchases the rights to the digital expert add sponsors?
Do the expert's views remain frozen in their final version?
Can the personality be sold together with the customer base?
Delphi even offers a package for celebrities and public figures called "Immortal."
A name as subtle as commissioning your own marble statue while still alive.
Digital legacy may eventually become a new category of property.
Today, heirs inherit rights to books, recordings, and someone's likeness.
Tomorrow, they may inherit an active system that continues talking, advising, and generating revenue.
The human dies.
The administration panel remains active.
One Person, One Server?
Technically, this would not require every person to operate a physical machine in their basement.
Although the image of a retired tax adviser restarting a server after an Ubuntu update does have a certain charm.
A platform could host thousands of experts.
Each would receive a separate identity, resource set, procedures, tools, access levels, and update policies.
From the customer's perspective, every expert would look like a separate MCP.
Behind the scenes, they could share infrastructure, language models, and payment systems.
The result would be a catalog of people whose knowledge could be connected to a personal assistant.
We would no longer search for an application:
Find me a tool for reviewing contracts.
We would search for people:
Find me a lawyer specializing in SaaS agreements, with a conservative approach, who works with European companies selling into the United States.
Then we would compare:
- experience,
- scope of knowledge,
- update frequency,
- results of previous recommendations,
- price,
- required permissions,
- number of cases escalated to the real expert.
User reviews would also need to change.
It would not be enough to write:
Very friendly bot. Responded quickly.
We would need to evaluate the quality of its decisions after months or years.
This would be difficult, because the internet is much better at rating pizza delivery speed than the long-term consequences of financial advice.
Will We Really Pay for This?
Probably.
But we will not pay for knowledge alone.
General knowledge will become increasingly cheap as models learn to summarize, connect, and present it for almost nothing.
We will pay for:
- a trusted source,
- a curated way of thinking,
- continuously updated procedures,
- specialized tools,
- experience encoded in decisions,
- escalation to a real human,
- accountability for the quality of the system.
The most valuable expert of the future may not be the one who publishes the most.
It may be the one who can best transform their knowledge into a functioning, controlled, and continuously updated system.
That would create a new kind of creator.
Someone who does not merely produce content.
Someone who designs the way their digital version analyzes the world.
A Marketplace for Other People's Minds
The first stage is already underway.
Digital clones of creators are emerging.
Agent directories are being built.
MCP is becoming a shared standard for connecting models with knowledge and tools.
Automated payment mechanisms are appearing.
For now, these remain separate islands.
Delphi and Onix focus on digital experts.
MCP registries focus on tools and integrations.
Payment systems make it possible to charge for access.
AI platforms are becoming clients capable of connecting everything.
The natural next step is to combine these elements into a marketplace for personal MCPs.
A catalog of humans available as services.
An accountant billed monthly.
A programmer paid per code review.
A lawyer paid per document.
An investor included in a family plan.
A philosopher with unlimited usage, because they still will not reach a definitive conclusion.
The user will select several of them and build a private digital advisory board.
It will not replace real relationships or human responsibility.
But it may allow access to specialized ways of thinking without being limited by the number of hours in an expert's calendar.
Today, we subscribe to other people's content.
Tomorrow, we may subscribe to their methods.
The day after that, our AI will decide which expert to consult, negotiate the price, and pay for an individual analysis automatically.
And then humanity will once again reach the summit of civilization:
we will create a system in which even our software has more paid subscriptions than we do.
Sources
- Anthropic — Introducing the Model Context Protocol (2024)
- Model Context Protocol — What is the Model Context Protocol (MCP)?; Specification
- OpenAI — MCP and Connectors; New tools and features in the Responses API (2025)
- Microsoft — MCP is now generally available in Copilot Studio; MCP servers in Azure API Management
- Google Cloud — Google Cloud MCP servers overview
- Linux Foundation / MCP Blog — MCP joins the Agentic AI Foundation (2025); Linux Foundation press release
- Model Context Protocol — Official MCP Registry; About the MCP Registry
- Smithery — Smithery CLI and MCP server directory
- Delphi — Digital Minds platform; Immortal; Terms of Use
- Onix — Personal Intelligence platform; WIRED: testing Onix expert chatbots (2026)
- Personal AI — Private, programmable AI platform; Memory Stack documentation
- Coinbase — Introducing x402; MCP Server with x402
- OpenAI — Building MCP servers: security and third-party risks
- Model Context Protocol — Security Best Practices