Contents

1. Introduction
2. Energy Warfare In Ukraine
3. The United States And Energy Tensions
4. Energy Warfare In The Middle East
5. Europe Under Pressure
6. Concluding Remarks

Bitesize Edition

• Energy Infrastructure as a Target — Energy infrastructure is increasingly being used as a strategic target in both the Russia-Ukraine conflict and the Middle East. Despite U.S. efforts to halt attacks on energy facilities in Russia, strikes continue, raising the risk of further disruption to global energy supplies. The longer these conflicts persist, the greater the pressure on nations that rely on stable energy flows.
• The United States’ Energy Dilemma — Although the United States promotes its relative energy self-sufficiency, rising diesel prices and potential export restrictions reveal its exposure to global supply disruptions. Expanding access to Venezuelan crude could help address some longer-term pressures, but it is unlikely to ease immediate issues. The United States must balance domestic energy prices against the consequences of restricting supplies to international markets.
• Europe and Asia Under Pressure — Disruptions in the Strait of Hormuz are placing additional strain on energy-importing nations. Europe faces growing pressure to secure crude oil, refined products and natural gas ahead of winter, while competition for limited supplies could leave parts of Asia vulnerable, as occurred during COVID. Alternative routes and military protection may reduce some risks, but cannot guarantee uninterrupted energy flows.
• The Risk of Escalation — Nations may increasingly turn to strategic reserves, alternative suppliers, reduced domestic demand and export restrictions to protect their energy needs. While these measures can provide short-term relief, they may also intensify competition and contribute to further geopolitical tensions. Without de-escalation in Ukraine and the Middle East, efforts to secure energy supplies could create additional pressure on an already strained global system.

Introduction

We’ve seen key developments in many geopolitical narratives over the course of the last few weeks.

The source of the foiled attack on the UK’s Fairford base has seen some blame the Iranians. The UK notably stated last week that Iranian attacks on UK soil would not be tolerated.

At the UN, the United States and China extended their trade truce into January, which will raise the stakes of the G20 in Miami.

Conflict restarted in Ethiopia, with tensions in the Tigray region once again elevated.

And the security deal between the United States and Greenland was a scenario we flagged here back in February. The United States, in an optimal scenario, would want a strategy that was low-severity, fast-speed, and low-cost, in its efforts to boost relations with Greenland.

The Future of Greenland

Read full story

The United States seems to have chosen a strategy of long-term persuasion with Denmark and Greenland, rather than the riskier, escalatory scenarios that would have occurred at a faster speed. As such, the United States prioritised a low-severity and low-cost strategy to boost Greenland’s security.

Finally, in Libya, tensions continue to rise with recent drone attacks and force majeure at energy facilities.

There are almost too many developments to discuss. As a result, today, we’ll continue the discussion we started last week on the conflict in Ukraine, before returning to each of these topics above over the coming weeks.

For today, we’ll explore other developments related to the Ukraine conflict, specifically energy warfare. Energy infrastructure is becoming a strategic target and a method through which nations can hurt one another. This is also relevant to the Middle East. The effects of these disruptions are starting to have clear, global ramifications. Worringly, the longer these effects last and the deeper they become, the more likely nations could begin to change their behaviours in an effort to address these vulnerabilities.

Today we’ll explore where the United States is relevant to these issues, who is under the most pressure due to recent energy tensions around the world, and what actions they could take in the future. Let’s dive in below.

Energy Warfare In Ukraine

In the last few weeks, we’ve seen the United States urge both Ukraine and Russia to halt strikes on energy infrastructure.

Both sides initially agreed, and then continued to hit energy infrastructure anyway.

Last weekend, Ukraine hit the Yaroslavl Oil Refinery, and fire was visible at Kapotnya Oil Refinery in Moscow.

This week, we saw a Ukrainian drone hit Russia’s Novoshakhtinsk refinery.

And so, despite the United States pushing for an end to these attacks on energy infrastructure, the Ukrainians aren’t stopping their attacks as of today.

Zelenskyy stated he will end strikes if Russia makes concessions, but Russia has continued to stick to its brief from 2022’s Istanbul Communiqué, and likely won’t shift stance after four and a half years without a change. It seems the Ukrainians will hence continue these attacks on refineries because a pursuit of peace does seem difficult today.

It is possible that the United States could increase the pressure on Ukraine to halt these strikes on energy infrastructure. They could refuse to send support, or cut intelligence support for Ukraine, for example.

Intelligence support has been another topic of discussion, because it’s been reported that the Russians have been supplying Iran with satellite intelligence to hit U.S. bases in the Middle East. If the United States cut intelligence support for Ukraine, could the Russians do the same for Iran in return, which could actually benefit the United States? This is speculative, but it’s worth considering because it highlights how individual incentives can often override relations with other nations.

Alternatively, we could see Trump and the United States accept that Ukrainian strikes on energy infrastructure won’t stop, which would likely spell a continuation of this conflict and all of these new, escalatory strategies that both the Russians and Ukrainians are pursuing.

Trump has somewhat increased pressure recently on Russia, with the sanctions bill being signed into law by Trump on September 18th, but this likely will have no effect on the conflict between Russia and Ukraine, and these current energy strikes.

However, the wider effects of global energy supply chain disruptions are having an effect on the United States.

The United States And Energy Tensions

In the last few weeks, the Trump Administration has been promoting the United States’ relative self-sufficiency when it comes to energy. Without looking deeper, it could seem that the United States personally may not need Ukraine to stop attacks on energy infrastructure, or need to address the wider energy disruptions around the world at all.

Despite this recent narrative, diesel prices in the United States have hit a record national average of $6.50 in the last week. But the Trump Administration continues to portray itself as relatively immune to the current disruptions to global energy supply chains.

In direct contrast to this, it has been reported by some that the United States is considering a diesel export ban.

If the United States does restrict diesel exports, this reduces supply in international markets, which can hurt other nations.

Alternatively, maintaining or increasing exports could risk domestic prices in the United States rising further. A boost of exports seems especially unlikely before the midterms, when Trump’s approval rating is falling, and he may want to offer Americans some economic relief from rising diesel prices at the pump.

Also, despite the United States painting itself as immune to energy disruptions, in the background, the United States is making moves to address these issues. A few weeks ago, the United States announced its intention to expand its presence in Venezuela through Chevron. Venezuelan crude is heavier crude, which U.S. refineries are well suited to handle, and that eventually becomes diesel when refined. However, any moves made in Venezuela will be slow and won’t address any short-term strain in energy markets.

Another nation with heavier crude is Canada, but due to the current trade spat with the United States, increased flow of Canadian crude to the United States looks unlikely.

These disruptions are also multifaceted. A nation can have access to crude oil through domestic production, reserves, or imports, for example. But products, such as diesel, need to be refined. Whether these refined products can reach where they need to be is one of the strongest aspects contributing to this current issue.

Of course, energy warfare isn’t only relevant to the conflict in Ukraine, but also to the Middle East, where the United States is currently embroiled in war with Iran. This has led to disruption in the Strait of Hormuz and damage to energy infrastructure across the Middle East.

Energy Warfare In The Middle East

In the last few weeks, we saw strikes hit the East-West Pipeline, which were blamed on Iran-backed militias.

This pipeline is vital for Saudi Arabia, with it rerouting 5 million barrels per day of crude oil from the Strait of Hormuz to Yanbu on the Red Sea. As one of the key pipelines providing relief from the disruptions in Hormuz, its continued functioning is necessary if energy tensions aren’t to worsen.

The pipeline is now back up and running. However, tensions with the Houthis remain elevated, and similar disruptions could arise once again. The worry is whether these disruptions remain temporary, which can be managed, or if sustained damage could see the pipeline stop operations completely.

In the Middle East, escalation remains a scenario that could occur, with Trump rejecting Iran’s deal at the UN and stating he will restart strikes after the midterm elections in early November. Iran and the Houthis may not wait until after the midterms to pressure the United States and the wider Middle East region. These threats could see the Houthis once again target the East-West pipeline, and this time, they may hit the pipeline so flows cannot be restarted. The remains a risk that nations of the world need to consider, and need to prepare for.

The longer both of these conflicts continue, the worse these effects could be. This could see us reach a scenario where the negative effects of global energy supply chains are being disrupted so heavily that nations begin to take action to address these issues. The worry is that many of these actions could increase global tensions.

Despite some pressure being seen in the United States when it comes to these energy disruptions, the much bigger strain is in Europe.

Europe Under Pressure

In the last week, the UK has offered Saudi Arabia defensive support, and France is sending troops to Yanbu. This could mark the beginning of nations pursuing actions to address the current energy disruptions.

Saudi Arabia has recently stated that no crude oil was sent to Europe in September, and that the same will occur in October. In general, Europe is incredibly reliant on energy imports, despite differing energy profiles for different nations.

With winter coming soon, Europe is also typically buying natural gas around this time to fill its gas storage. Europe is also reliant on imports for natural gas, which is also disrupted by restrictions in the Strait of Hormuz. The Strait is responsible for 20% of global trade of liquefied natural gas.

The longer the Middle East conflict continues, the more restricted flows into Europe could be, and the worse the financial effects will become. Also, Europe continues to reject Russian energy sources due to the conflict with Ukraine, and they’re not going to shift stance on this. Hence, the Europeans could begin to make moves to ensure energy flows continue to Europe. Otherwise, petrol and diesel prices will continue to rise.

This could see Europe compete financially for crude oil imports and the import of refined products. This impact would especially be felt in Asia, despite the fact that much of Russia’s energy flows are now directed towards Asia.

In the past, the nations that lose out in these environments are those that are outbid on energy shipments. During COVID, it was nations such as Pakistan and Bangladesh that were outbid by European nations, and these Asian nations underwent energy shortages and rising inflation.

Today, the restrictions in the Middle East will worsen the longer they continue. Also, if Russian flows are restricted by Ukrainian drone strikes, this will further restrict energy flows, but this will especially hurt Asia due to its dependence on imports.

Finally, since the United States is also seeing prices spike in diesel, expecting U.S. exports to help other nations cannot be assumed.

This is creating a global energy environment that is suffering, and that has the potential to suffer even more.

Concluding Remarks

The world is reaching a moment where the underlying effects of rising geopolitical tensions are growing more visible.

This strain on our global energy system, on financial systems through rising inflationary pressures, and on the potential for trade restrictions is turning into a problem for many geopolitical actors.

If we continue down this current path without efforts to de-escalate in Ukraine and Iran, nations that are vulnerable to energy disruptions and the second-order effects of this could be more likely to make moves to address this.

These moves could be unilateral, such as reducing domestic demand, releasing strategic reserves, or export restrictions. But these moves can also raise tensions on a multilateral level, especially export restrictions. Many strategies to secure energy seem like they could raise tensions even further.

Also, if a nation pursues an alternative trade route to receive supplies, this still contributes to economic strain and inflationary pressures because alternative supply routes take longer.

Also, military protection is now being seen as a potential strategy, with the UK and France offering help to Saudi Arabia. But this doesn’t stop a drone strike from hitting a pipeline, or a vessel from being hit by a projectile.

If such strategies don’t work to address energy disruptions, more strategies will be required. As such, nations could be incentivised to pursue strategies that are escalatory in an effort to immediately address issues.

This is why de-escalation in Ukraine or Iran may be necessary to avoid rising tensions in global energy markets. Without de-escalation, Russia and Middle Eastern disruptions are limiting some of the biggest players in global energy markets. The longer these conflicts continue, the greater the impact these disruptions will have, and the more likely nations are to need to take action to address these issues.

Do you believe we will see escalation as nations move to secure limited energy supplies? Or will the geopolitical actors involved in conflict and energy restrictions push for peace, recognising that they too will suffer in a world that continues to escalate even further?

Thanks for reading! I’d greatly appreciate it if you were to like or share this post with others! If you want more, then subscribe on Substack for these posts directly to your email inbox. I research history, geopolitics, and financial markets to understand the world and the people around us. If any of my work helps you be more prepared and ease your mind, that’s great. If you like what you read, please share this with others.

Key Links

Geopolitics Database

The Geopolitics Explained Podcast

The Geopolitics Explained Reading List

If you want to see daily updates and discover other newsletters that suit you, download the Substack App.

Get more from Dylan Muggleton in the Substack app

Available for iOS and Android

Get the app

You can become a paid subscriber to support my work. There are paid posts every Thursday and long-form monthly articles in my global questions series exclusively for paid subscribers. The Geopolitics Database is also accessible. Read Geopolitics Explained for 20p per day or start a free trial below to find out if my work is for you! I appreciate your support!

Energy Warfare Takes Centre Stage was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.

Claude can explain crypto, summarize market narratives, and reason through investment questions surprisingly well.

But crypto moves too quickly for static knowledge to be enough.

Prices change around the clock. News breaks continuously. Market sentiment shifts. An altcoin can move sharply within hours, while an AI model relying only on previously learned information may have no idea what just happened.

So I wanted to test what changes when Claude no longer has to rely on static knowledge for the financial facts behind its analysis.

For this experiment, I connected Claude directly to Financial Modeling Prep through its MCP server and used NEAR as the test asset.

I then gave Claude a series of practical research tasks and evaluated how well it could retrieve, connect, and summarize the evidence.

The experiment was intentionally simple:

• Check what NEAR is doing in the market right now
• Put the recent price movement into historical context
• Find relevant recent developments and news
• Combine the evidence into a concise research brief

No large crypto screen. No comparison across ten tokens. No attempt to build an automated trading bot.

The goal was simply to see whether connecting Claude to live financial data through MCP could make crypto research faster, more current, and more useful.

In other words, instead of asking Claude to remember what it knows about NEAR, we are asking it to retrieve the current evidence first and reason from that evidence.

That is the experiment.

Let’s see how it performs.

The Setup: Claude + FMP MCP

The setup is simple:

Claude → FMP MCP Server → Live Crypto Data

Financial Modeling Prep supports MCP, which means Claude can discover and query FMP’s available financial tools directly instead of requiring us to manually call APIs, parse JSON, and paste the results into a prompt. FMP’s crypto coverage includes real-time quotes, historical price data, and cryptocurrency news, which is enough for the experiment in this article.

To connect FMP in Claude:

1. Open Settings → Connectors
2. Select Add custom connector
3. Paste the FMP MCP URL:

https://financialmodelingprep.com/mcp?apikey=YOUR_FMP_API_KEY

4. Replace YOUR_FMP_API_KEY with your active FMP API key

5. Start a new Claude conversation

Once connected, Claude can discover the FMP tools available through the MCP server and use them when responding to the prompts we send.

For this article, we will keep the experiment focused on NEAR and test four things:

current market data → recent price context → relevant news → a concise research brief

AI Agent (MCP Server) - FMP

The point is not to build a trading system. It is to see how much more useful Claude becomes when the facts behind the analysis come from a current financial data source.

Test 1: What Is NEAR Doing Right Now?

Let’s start with the most basic research task: understanding what NEAR is doing in the market right now.

Normally, this would mean opening a crypto-price website, checking the current price, looking at the recent move, switching to a historical chart, and manually putting the numbers into context.

With FMP connected through MCP, Claude can retrieve that information directly.

For this first test, I asked Claude to:

• Confirm the correct NEAR/USD symbol available through FMP
• Retrieve the latest price, percentage change, trading volume, and daily range
• Calculate the 7-day and 30-day price change
• Compare the latest move with NEAR’s recent 30-day price behavior

The goal here is not to predict NEAR’s next move. It is simply to see whether Claude can retrieve current market data, perform the calculations, and turn the raw numbers into useful market context without any manual data collection.

Using only data available through the connected Financial Modeling Prep MCP server, analyze the current market position of NEAR Protocol.

1. First, identify and confirm the correct NEAR/USD cryptocurrency symbol available in FMP.
2. Retrieve the latest available crypto quote for NEAR and report:
   - Price
   - Latest percentage change
   - Trading volume
   - Day high and low, if available
   - Timestamp or data date
3. Retrieve historical daily prices and calculate:
   - 7-day price change
   - 30-day price change
   - Highest and lowest closing price over the last 30 days
4. Briefly explain whether the latest move appears small, moderate, or unusually large relative to NEAR's recent 30-day price behavior.

Use only FMP data. Do not use web search or outside sources.

Do not give a buy, sell, long, or short recommendation.

Keep the response concise and return:
- A small market snapshot table
- 3–5 bullet points explaining the recent price context

What Test 1 Showed

The first test already shows the difference between asking Claude about a cryptocurrency and letting Claude retrieve the market data behind its answer.

Claude correctly identified NEARUSD as the FMP symbol and combined the latest quote with historical prices to provide both short-term and 30-day context.

At the time of the run, NEAR was trading around $4.94, down 8.18% on the latest quote. Looking only at that daily move could make the decline appear unusually severe.

The historical data changed the picture.

NEAR was still up roughly 16% over seven days and more than 160% over the previous 30 days according to the FMP data used by Claude. The historical series also showed several double-digit daily moves during the month.

That allowed Claude to place the latest decline in context: it was a relatively large daily move, but not outside the range of volatility NEAR had already experienced during the recent rally.

The useful part here is not the price itself. Any crypto website can show that.

The value comes from Claude being able to retrieve the current quote, pull the relevant history, calculate the changes, and summarize the market context in one interaction — without us manually collecting and pasting the numbers first.

Test 2: What Changed Around NEAR?

The first test established what NEAR was doing in the market.

The next question is more useful:

What changed around NEAR while this move was happening?

Crypto news can be noisy. A single token may appear in exchange announcements, ecosystem updates, partnership stories, market commentary, and speculative headlines within the same week.

Rather than reading everything manually, I asked Claude to use FMP’s crypto news data to identify the developments most relevant to NEAR’s recent move.

The goal was not to force an explanation for the rally.

Instead, I wanted Claude to:

• Retrieve recent NEAR-related news from FMP
• Remove duplicate or low-value headlines
• Identify the developments that appear most relevant
• Place those developments on the same timeline as NEAR’s recent price movement
• Clearly separate confirmed facts from possible interpretation

This is an important distinction. A headline appearing near a price move does not prove that it caused the move.

Using only data available through the connected Financial Modeling Prep MCP server, review recent developments related to NEAR Protocol.

1. Confirm the NEAR/USD symbol used by FMP.
2. Retrieve NEAR-related cryptocurrency news from the most recent 14 days.
3. Remove obvious duplicate stories and low-value mentions.
4. Select up to 5 developments that appear most relevant to NEAR itself.
5. For each development, report:
   - Date
   - Headline
   - Source
   - Short summary
   - Why it may matter for NEAR
6. Compare the timing of these developments with NEAR's recent price movement from FMP historical data.

Important rules:
- Use only FMP data.
- Do not claim that any news item caused the price move unless the evidence clearly establishes that relationship.
- Clearly distinguish confirmed facts from interpretation.
- Do not give buy, sell, long, or short recommendations.

Return:
- A compact table of the most relevant developments
- A short timeline summary explaining what changed around NEAR during the recent move

What Test 2 Showed

This test was more useful than simply asking Claude to summarize crypto news.

Claude retrieved a large number of NEAR-related headlines, recognized that many of them were repetitive price alerts, and reduced the noise to a smaller set of developments that were actually relevant to the project.

Several meaningful events appeared during the same period:

• NEAR Intents crossed roughly $30 billion in cumulative volume
• Ondo brought tokenized U.S. stocks and ETFs to near.com and NEAR Intents
• Hyperliquid added NEAR/USDC spot trading
• Bitwise’s proposed NEAR ETF moved through important SEC and NYSE Arca registration steps
• NEAR introduced private limit orders for cross-chain assets

The timeline also showed why this kind of analysis needs caution.

The strongest phase of NEAR’s price advance occurred around the same time that ETF-related coverage intensified, but that timing alone does not prove the ETF news caused the rally. Several ecosystem developments appeared within the same few days, while the broader crypto market was also moving.

Claude handled this reasonably well by separating temporal overlap from confirmed causation.

That is one of the more useful things an MCP-connected workflow can do: retrieve the relevant headlines, remove obvious noise, align events with price data, and leave the final interpretation open where the evidence is not strong enough.

Test 3: Can Claude Turn the Evidence Into a Research View?

By this point, Claude has already retrieved NEAR’s recent price behavior and the developments reported around the same period.

The next test is harder:

Can it turn that evidence into a useful research view?

This is where an AI assistant becomes more interesting than a simple data-retrieval interface.

Instead of asking for another list of prices or headlines, I asked Claude to combine the available FMP evidence into three things:

• The strongest factors supporting the current NEAR narrative
• The main risks or reasons for caution
• The developments worth monitoring next

The goal is not to generate a price target or decide whether NEAR should be bought or sold.

It is to see whether Claude can move from retrieving information to organizing an investment-research question around evidence.

Using only data available through the connected Financial Modeling Prep MCP server, build a concise research view for NEAR Protocol based on the latest available market data and recent developments.

Use the recent price history, crypto news, and other relevant FMP evidence available for NEAR.

Structure the response into four sections:

### 1. Current Market Context
Summarize NEAR's recent price trend and volatility using the latest available FMP data.

### 2. Factors Supporting the Current Narrative
Identify up to 4 developments that could reasonably support investor interest in NEAR.

For each one:
- State the confirmed fact
- Explain why it may matter
- Do not assume it caused the price move

### 3. Risks and Reasons for Caution
Identify up to 4 risks or uncertainties supported by the available evidence.

Consider:
- Recent price volatility
- Whether the rally has been unusually fast
- Uncertainty around the financial impact of recent developments
- Whether important claims remain unverified or incomplete

Do not introduce risks that are not supported by the FMP evidence.

### 4. What to Watch Next
List up to 5 specific things that would be useful to monitor over the coming days or weeks.

Focus on observable items such as:
- Follow-up ETF developments
- Trading or liquidity changes
- New ecosystem announcements
- Changes in price behavior
- New NEAR-related news

Important rules:
- Use only FMP MCP data.
- Clearly separate confirmed facts from interpretation.
- Do not provide a price target.
- Do not predict that NEAR will rise or fall.
- Do not provide buy, sell, long, or short recommendations.
- Keep the output concise and evidence-based.

What Test 3 Showed

This was the point where Claude moved beyond data retrieval.

Instead of returning another collection of prices and headlines, it combined the evidence into a structured research view: what currently supports the NEAR narrative, what remains uncertain, and what deserves monitoring next.

The distinction between fact and interpretation was especially useful.

For example, Claude could confirm that the Bitwise NEAR ETF had moved through important SEC and NYSE Arca registration steps. But it also recognized that the available evidence did not yet provide a confirmed first trading date.

Similarly, the NEAR Intents volume milestone and Ondo integration were real developments, but the data available in this experiment did not establish their direct financial impact on the NEAR ecosystem.

That is a much more useful output than simply labeling every development “bullish.”

Claude was able to organize the research into three buckets:

What we know → What it might mean → What we still need to verify

And importantly, it did this without turning the result into a price prediction or trading recommendation.

For me, this is where the MCP connection becomes most useful. FMP supplies the current evidence, while Claude helps turn that evidence into a research process.

Want to Try It Yourself? Use This Prompt

After running the three tests separately, the pattern is fairly clear.

Claude works best when the prompt does three things:

1. Tells it exactly which asset to research
2. Restricts the analysis to current FMP MCP data
3. Separates confirmed facts from interpretation

If you want to repeat the same workflow for another supported cryptocurrency, you can use a prompt like this:

Using only data available through the connected Financial Modeling Prep MCP server, research [CRYPTO ASSET].

Confirm the correct USD symbol available in FMP.

Retrieve the latest quote and summarize:

Current price

Latest percentage change

Trading volume

Day high and low, if available

Retrieve historical prices and calculate:

7-day change

30-day change

30-day high and low

Retrieve relevant crypto news from the most recent 14 days.

Remove duplicate or low-value headlines and identify the most important developments.

Compare the timing of those developments with the recent price movement, but do not assume causation.

Build a concise research view with:

Current market context

Factors supporting the current narrative

Risks and uncertainties

What to watch next

Important rules:

Use only FMP MCP data.

Clearly separate confirmed facts from interpretation.

Do not invent missing information.

Do not provide a price target.

Do not give buy, sell, long, or short recommendations.

Keep the response concise and evidence-based.

The asset can be changed without rewriting the entire workflow.

For example, replacing NEAR with another supported cryptocurrency lets Claude repeat the same process: retrieve current data, add historical context, filter recent developments, and organize the evidence into a research view.

The prompt is not designed to predict the next breakout.

It is designed to make the first stage of crypto research faster and more structured.

AI Agent (MCP Server) - FMP

Conclusion

This experiment started with a simple question: what changes when Claude can access current financial data instead of relying only on what it already knows?

Using NEAR as the test asset, Claude was able to move through three stages of research:

retrieve the market data → investigate recent developments → synthesize the evidence into a research view

The interesting part was not any single number or headline. It was the workflow itself.

With FMP connected through MCP, Claude could retrieve the relevant data, filter out noise, add historical context, and organize the findings through natural-language prompts.

That makes MCP especially interesting for financial research.

Instead of manually collecting data first and asking the model to analyze it afterward, the research process becomes much more direct:

ask the question → retrieve the evidence → analyze the context

And NEAR was only one example.

The same approach can be extended across other supported cryptocurrencies, equities, financial statements, earnings data, news, and other datasets available through FMP.

As MCP-connected tools continue to expand, the more interesting question may no longer be simply what Claude knows.

It may be:

What data can Claude reach when you ask the question?

I Gave Claude Live Stock Market Data Through MCP — Here’s What It Could Do in 10 Minutes was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.

Have you ever found yourself in a situation when your software needed a language model to decide upon trivial things?

A customer email arrives. Which department should process the query?

A task fails. Should the agent retry?

A potential customer arrives. Does he or she deserve to talk to a sales person?

A message needs to be moderated. Is the content safe?

Those questions are not “writing” ones.

Those are decision problems.

This is the kind of tasks for which some new models have been developed recently such as TypeSafe’s Jev.

Now there is an intriguing open-source solution for those: Laya, a non-autoregressive System 1 decision model from Convai Innovations that allows one to run inference locally without hitting the hosted API.

Laya is designed to solve the same type of problems: provide some application state to a model, ask structured questions and get machine-readable decisions as responses instead of writing a lot of generated text. The current checkpoint family of the project consists of English, multilingual and typed-decision weights released under the Apache 2.0 license.

That seems to be straightforward.

But this is a useful shift in the way one can build AI-powered applications.

The problem: We are using giant models to solve trivial decision problems

Imagine an agent processing 10,000 emails. For each one of them, you can ask a large language model to:

“Please read the email and make a decision regarding the department that should handle the case.”

And receive such response:

“Looks like a billing-related matter since the customer has said…”

It is easy for a human to understand that answer. However, your software needs:

department = billing
confidence = 0.82

The difference is huge.

LLMs generate text token by token. Your application will have to parse it before being able to proceed.

Laya works differently. It is not supposed to write an answer understandable for a human being. This model is intended to make structured decisions that the other part of the application can act upon. The project’s documentation defines it as a non-autoregressive decision model producing typed answers with probabilities in one forward pass.

The main idea is:

LLMs generate.
Decision models decide.

Sometimes, the latter one is what you need.

What is Laya?

Laya takes something like email, support ticket, document, JSON object or some other piece of application state and asks predefined questions about it.

The current Laya project has three primitive decision types:

Choice — select an option from a predefined list.

Score — put something on the ordered scale.

Noul — evaluate the probability that the given yes/no statement is true.

These are the same primitive decisions used in Jev, that is why Laya can be regarded as an open-source alternative in the same architectural category.

For example:

Input:
"We were charged twice for March."

Choice:
billing — 0.91
technical — 0.06
sales — 0.03

Score:
urgency = 8.2 / 10

Noul:
customer threatens to cancel = 0.14

Your application can then decide what to do based on those scores.

That is the key point here: Laya decides. Your application executes business logic.

Why this model can be faster

According to the Laya GitHub benchmarks, its English checkpoint was measured at about 39.5 ms for a single question and its multilingual checkpoint at 32.8 ms on a Tesla T4 GPU. Multiple questions can be batched and thus processed in one inference pass, reducing average time even more.

The project makes a comparison with independently published measurements of Jev: roughly 236–276 ms p50 for a single question, noting at the same time that the comparison was performed using different hardware, prompts and measuring setup.

The given transcript makes the same point from a different angle: Laya can make several structured decisions based on the same piece of text in one inference pass, making it highly useful in workflows with lots of little decisions to make.

Thus, the advantage of Laya is not just “smaller AI”. It is AI optimized for a particular problem.

Practical example: Automate lead routing

Imagine that you are running a SaaS company. A potential client writes:

“We’re a 200-people company and want to automate our customer support workflow. Please, contact us regarding the pricing.”

Instead of passing that message to the powerful LLM and asking it to decide everything, you could ask Laya three questions:

1. What is the type of the customer?
   → Enterprise / SMB / Individual

2. How strong is the client's buying intent?
   → Score 1–10

3. Should the sales person contact the client?
   → Yes probability

Let’s imagine that the result is:

customer_type = enterprise
intent = 8.6
follow_up = 0.91

Then the usual application code can decide what to do next:

if enterprise
and intent > 7
and follow_up > 0.85:

    send_to_sales()

That is the power of this approach.

You are not asking an LLM to run the whole application. You are using a dedicated model as a decision layer within the application.

You can run it yourself

This is one of the key differences of Laya from a hosted decision API.

The current project is distributed under the Apache 2.0 license and provides local usage of the laya Python package according to the documentation.

A sample application using Laya looks as follows:

import laya

agent = laya.load("convaiinnovations/laya")

result = agent.predict(
    {
        "subject": "Duplicate charge",
        "body": "We were billed twice. Please refund the duplicate."
    },
    {
        "department": {
            "type": "choice",
            "instructions": "Which team should handle this?",
            "criteria": {
                "billing": "invoices, payments, refunds",
                "technical": "bugs and outages",
                "sales": "pricing"
            }
        },
        "urgency": {
            "type": "score",
            "instructions": "How urgent is this?",
            "criteria": ["not urgent", "soon", "blocking"]
        },
        "churn_risk": {
            "type": "noul",
            "instructions": "Does the user threaten to cancel?"
        }
    }
)

This architecture is significantly different from the standard chatbot usage case.

You define the state. You define the decisions you need.

Laya returns the structured results. Your application decides what to do next.

Laya gets really interesting

There are several scenarios where it can be applied.

Model routing

Imagine that your application has three models:

Small model → cheap
Medium model → balanced
Large model → expensive

Instead of always using the expensive one, you can let Laya decide:

Is this request simple?

YES → small model

NO → large model

Model routing is specifically highlighted as an application scenario of this type of decision model in the given transcript.

Browser agents

A browser agent has to make decisions about:

Which button to click?

Where to scroll to?

Is this the right element to interact with?

Those decisions have to be made repeatedly.

Instead of using the general purpose reasoning model for all those clicks, one can use a dedicated decision model, potentially gaining significant performance boost.

Guardrails and moderation

The same approach can be used for:

Is this content unsafe?

Should this request be escalated?

Does this tool call seem suspicious?

Should this agent retry?

The key thing in all those cases is the following algorithm:

State → Decision → Software action

But Laya is not magic

That is where the current project’s documentation becomes especially helpful.

The best typed-decision score published for Laya belongs to its fine-tuned laya-typed-decisions checkpoint. All the other Laya models are significantly inferior for that specific benchmark. The project explicitly describes Laya as a fast base model that can be specialized and does not imply that it will work well for every decision-making task automatically.

There is also a limitation related to large choice sets.

According to the current documentation, the project recommends to keep choice decisions with no more than 20 options, as the model has a fixed token budget shared across options and thus will be unable to make a decision in case of too big choice set.

Instead of:

Choose one of 100 categories

it is better to design:

Step 1:
Which broad category?

Step 2:
Which specific category?

It is actually a better software design in most of the cases.

Also, the project recommends using multilingual checkpoint outside English and warns about the necessity of testing the calibration of probability scores on your own data before using them for automation.

So, should you use Laya?

The important question is not:

“Is Laya better than Jev?”

It is:

“Do I need a hosted decision API or want to own my decision layer?”

Jev gives you hosted solution. Laya provides open weights that can be run and integrated directly.

Thus, the decision is more about deployment, ownership, cost, customization, latency and the quality needed for your specific application, not about AI model A vs AI model B.

For developers building agents, automations, routing, moderation, browser control and classifiers, that is an interesting new tool.

The bigger idea is even more important. We may see the future where a single model does not do everything.

One model generates text.

Another reasons.

Another retrieves the information.

And yet another just answers a tiny question fast enough for software to act on it thousands of times per second.

That is the reason to pay attention to Laya. Not because it replaces LLM.

Because it implies that we do not need LLM for every decision.

Laya: The Free, Open-Source Alternative to Jev for AI Decisions was originally published in Predict on Medium, where people are continuing the conversation by highlighting and responding to this story.