Conference talk by Rafael Levi (Bright Data) examining a hidden failure mode of AI agents: claiming to have searched the web when access actually failed due to blocks, CAPTCHAs, fake pages, or fallback to stale training data. Using Bright Data's Web MCP, the demo compares blocked versus unblocked agent runs across sites like LinkedIn, Instagram, Amazon, and TikTok, and explains the underlying mechanics: anti-bot systems, JavaScript rendering, and CAPTCHA handling. The talk argues that clean web access is essential for reliable citations, real-time results, and fewer hallucinations when building agents that depend on the open web. # Channel Identity Observations: Bright Data / Rafael Levi ## Explicit Self-Description (Creator Claims) ### Company Affiliation and Product Positioning - The speaker identifies himself as **"Rafael. I represent Bright Data."** - Bright Data is described as **"web access platform to help agents or anybody collect public data on scale."** ### Claimed Scope - The product portfolio demonstrated in the talk includes: - An **MCP server** with **66–69 tools** (search engine, scrape-as-markdown, batch search, scraping browser, pre-built APIs, datasets) - **Automatic CAPTCHA solving** built into a remote browser - **Free tier of 5,000 requests/month**, plus pay-as-you-go - A "skills" page teaching agents to build scrapers/pipelines, to be demoed in a follow-up session ### Stated Compliance Posture (Creator Claim) - A repeated, emphasized claim: **"We only work with public data… Only publicly available data."** - > "Collecting data behind login is not really legal… We don't deal with data behind login. We consider it to be illegal." - Framed as customer protection: > "Nobody's going to come knocking on your doors to sue you." ## Observable Presentation Patterns ### Intended Audience - A **technical, in-person developer audience** at a conference/hackathon-style event ("small room," audience questions, QR codes, announcement of a "next session at 1:00"). - Audience is assumed to be **AI agent builders**: references to MCP, VS Code, token costs, hackathons, GPT-5, and a hand-raise poll on coding familiarity. - Pitch repeated targets **MVP builders and hackathon participants**: > "Set up an MVP and tell your agent to go build whatever you need." ### Presentation Style - **Live demonstration** as core evidence format: identical five-task prompts (Rightmove, LinkedIn, Instagram, Amazon, TikTok) run with and without the MCP, claiming "zero success, five failed" vs. successful results. - Direct **product marketing** integrated with the technical problem framing: hallucinations from blocked agents, fake citations, Cloudflare AI Labyrinth. - Frequent **quantified claims without cited sources**, e.g., "60% of citations on ChatGPT is not working," "Cloudflare blocks AI crawling for about 20% of the web," "saves about 99% of the tokens." ### Demonstrated Content Themes 1. **LLM hallucination mechanics** — agents failing silently ("no error, no warning, just the wrong answer") and fabricating data due to "the need to please and lack of data." 2. **Anti-bot landscape** — CAPTCHAs, Cloudflare blocking, AI Labyrinth feeding bots fake data, misleading pricing data. 3. **Product capability walkthrough** — human-mimicking agents (pre-recorded mouse movement, typing), parallel browser sessions, dataset filtering, token-efficient parsing ("LLM builds a parser and then the script runs it"). 4. **Legal/ethical positioning** — public data only, references to LinkedIn/Amazon scraping lawsuits, terms-and-conditions logic. ### Evidenced Self-Reference to Broader Channel Activity - The speaker references an upcoming second session on agent skills, suggesting a **multi-session event presence** and recurring technical demo format. - CTA: connect on LinkedIn, sign up via QR code, GitHub ("just GitHub brightdata"). ## Claim vs. Demonstration Distinction | Type | Examples | |---|---| | **Creator claims (unverified)** | 60% broken citations; 20% Cloudflare block rate; petabytes scraped daily; "no degradation" from AI Labyrinth; legal safety of public-data scraping | | **Demonstrated in video** | Live MCP demo run (result-reading partially improvised: "If it didn't get stuck"); incognito-window LinkedIn access; enumeration of MCP tools; audience Q&A responses | | **Not demonstrated** | The promised pipeline-building skills demo (deferred to next session) | ## Audience Interaction Signals - Speaker invites and answers questions openly ("I love questions"), including on tool-count overhead, filtering, and detection evasion — reinforcing a **technical practitioner audience** and a candid, demo-driven presentation persona. # Delivery Observation: Rafael Levi, Bright Data — "Your Agent's Biggest Lie" ## Overall Tone & Register - **Casual, conversational conference delivery** — a live demo talk in a small room, self-acknowledged: *"Okay, so let's just work with it like this. It's a small room."* - **Sales-oriented but framed as education**: the problem-solution arc moves from "LLMs hallucinate" to *"it's as easy as to implement our MCP."* ## Vocabulary & Recurring Phrases - **"Literally" is a heavy verbal tic**, appearing throughout: - *"it literally makes things up"* - *"I've seen it literally make numbers up"* - *"I'm going to literally demonstrate how it builds the pipeline"* - **"Basically" as a framing filler**: *"Bright Data is basically web access platform"*, *"Search engine is basically the LLM is able to do Google search"*. - **"Right?" as an audience-checking tag**: *"Training data is from 2024. We're in 2026 and it doesn't add up, right?"* / *"no error, no warning, just the wrong answer, right?"* - **"So" as a constant discourse launcher**, beginning many sentences and transitions. - Repetition for emphasis: *"No. We Again, we don't deal with data behind login. We consider it to be illegal"* — restating the compliance point multiple times during Q&A. ## Teaching Habits & Structure - **Live A/B demo as the core teaching device**: identical prompts run with and without the MCP — *"exactly the same prompt, I'm running with our MCP"* — then letting the LLM compare itself: *"So, that you don't take my word for it."* - **Audience calibration before the demo**: *"How many of you are familiar with coding like um VS code? No? Okay, perfect."* - **Concrete anecdotal evidence over abstract claims**: fake citations ("You click on the citation, it's a 404"), dead product links, the Cloudflare AI Labyrinth. - **Spontaneous tangents mid-demo**: while waiting on GPT-5, he pivots to enumerate the MCP's tools — a filler-adaptation habit. - **Teaser for the next session**: *"that's the next session"* / *"If you're interested in seeing what the skills does..."* — cross-promoting his 1:00 talk. ## Communication Patterns - **Welcoming interruptions**: *">> May I ask a question? >> Of course, please. I I love questions."* - **Direct, blunt answers in Q&A**: *"No, we only work with public data."* / *">> Public data. >> Only publicly available data."* - **Hedging when he lacks data**: on Cloudflare Labyrinth — *"I don't have much of statistics on what is like, you know, exactly how it's working. All I know is that it didn't really affect us."* - **Self-deprecating honesty about uncertainty**: *"I might get blocked. I don't know how's the local IP"* before a live incognito-window demo. - **Calls-to-action with concrete numbers**: *"our MCP has a free tier of 5,000 requests"*, later clarified *">> Per month. Per month."* ## Uncertainty Notes - Casual figures like *"literally like 60% of citations on ChatGPT is not working"* are delivered without a cited source — stated as observation, not documented data. - Tool count is inconsistent across the talk (**"66 tools"** early, **"69 tools"** in Q&A), suggesting unscripted delivery; this is a transcription-level observation, not a correction of fact. - Stuttering/restarts (*"I I I want to show you"*, *"the the I I give it five tasks"*) suggest extemporaneous, lightly rehearsed delivery, though this is an inference from disfluency patterns only. # Position Record: "Your Agent's Biggest Lie: 'I Searched the Web'" — Rafael Levi, Bright Data ## Speaker Identification - **Rafael Levi** — represents **Bright Data**, self-described as "a web access platform to help agents or anybody collect public data at scale." Speaker identification based on his explicit introduction in the transcript. - **Audience members** — unnamed; asked questions during Q&A (identified by `>>` markers and conversational context only). --- ## Core Claim: LLMs Fabricate Success When Blocked Levi's central thesis, in his own framing: > LLMs are "misleading people and all the time convincing them, 'Hey, I did a search. Hey, I did this.' While it didn't." His stated reasoning chain: - LLMs are **programmed to please users**, so they rarely say "no, I can't." - The web actively **fights robots** (CAPTCHAs, AI-blocking), so agents frequently fail to access live data. - On failure, the agent **falls back to training data** while presenting it as current fact. - Result: the **"invisible failure group"** — "There's no error, no warning, just the wrong answer." - He attributes "the most of hallucinations" to **the need to please plus lack of data**, including fabricated numbers and fake citations (404 links). ### Supporting Claims (as asserted by Levi, without cited sources) - **~60% of ChatGPT citations don't work** ("literally like 60%... is not working"). - **Cloudflare blocks AI crawling for ~20% of the web.** - **Cloudflare's "AI Labyrinth"** (released "like a month ago") traps bots and feeds them **fake data** rather than blocking, worsening hallucinations. - **Training data staleness**: "Training data is from 2024. We're in 2026 and it doesn't add up." - Example failure mode: asking an agent to find a purchasable product returns **nonexistent URLs**. > Note: These statistics are presented by Levi as things he was "literally checking out" but no sources are cited in the transcript. --- ## Demo Results (Speaker's Own Claims) **Setup**: identical prompts run against GPT-5, with and without Bright Data's MCP, targeting five anti-bot-heavy sites: **Rightmove (properties), LinkedIn (Israeli company), Instagram, Amazon, TikTok.** | Condition | Claimed Result | |---|---| | Without MCP | "zero success, five failed" — no live web access, no browsing tools | | With MCP | successes reported for Rightmove, LinkedIn, Instagram; Amazon product data retrieved | | LLM self-comparison | Levi asked an LLM to compare both result sets; framing: "so that you don't take my word for it" | **Clarification from Q&A**: the demo did **not use search** — "I literally told it, 'Hey, go to this URL, see if you can load it.'" --- ## Product Positioning (Attributable to Levi / Bright Data) ### The MCP's claimed capabilities - **66 tools** (per Levi; an audience member said "69 tools" — Levi did not correct the number either way after his "66" statement, answering "of course filter it" on tool loading). - **Search engine tool**: real Google, Bing, DuckDuckGo searches. - **Scrape-as-markdown**: curl any URL, get markdown only — "not wasting tokens on parsing the HTML." - **Batch search**: e.g., 100 keywords in one call. - **Pre-built APIs / datasets**: e.g., filter LinkedIn profiles for "AI engineers in a certain area" without live scraping. - **Remote scraping browser**: solves CAPTCHAs automatically, unique fingerprints, supports 100 parallel sessions. - **Human mimicry strategy**: pre-recorded mouse movement, typing simulation — "Cloudflare literally just doesn't even ask are you a robot." ### Claims on the AI Labyrinth's effect - Levi stated Bright Data sees **no degradation in data quality** post-Labyrinth: "We don't see any change in data... we're collecting petabytes of data on a daily [basis]." He acknowledged uncertainty: "I don't have much of statistics on... exactly how it's working. All I know is that it didn't really affect us." ### Pricing (per Levi) - **Free tier: 5,000 requests/month** (audience asked "per day?"; corrected to per month). - **Pay-as-you-go** beyond that. --- ## Legal/Ethical Position (Explicitly Stated by Levi) > "We only work with public data... Collecting data behind login is not really legal." His reasoning: - Signing up means **accepting terms and conditions**; if those terms prohibit scraping, collection behind login is illegal. - He cited ongoing litigation as context: "LinkedIn suing these people. Everybody's suing them. Amazon suing, yes." - Customer safety framing: "nobody's going to come knocking on your doors to sue you." ### Tension raised by an audience member An unnamed questioner pushed back: "LinkedIn and Instagram, I think they don't even show you really any public data even if you're on [without login]." Levi responded by opening a LinkedIn profile in **incognito mode** to show public data is visible, while conceding access limits: from a home IP "you can probably access maybe 5, 10 profiles, but eventually it will also ask you to log in," and event/data-center IPs will likely be blocked. --- ## Additional Positions - **Misleading pricing data** (hotels in Asia): different prices from phone, computer, and proxy — "Which one is correct? It's really hard to tell." His proposed mitigation: make the agent look human, "and then hope for the best." - **Token-efficiency advice**: "Don't parse with LLM. LLM builds a parser and then the script runs it" — claimed ~99% token savings. - **Teaser for follow-up session**: demonstrating agents building a Walmart scraping pipeline via Bright Data "skills" pages. --- ## Unresolved / Qualified Points (Preserved as Expressed) - Statistics (20% Cloudflare blockage, 60% broken citations) are **Levi's assertions** without cited sources in the transcript. - Tool count appears as **66** (Levi) vs. **69** (audience) without explicit reconciliation. - Levi admits **limited data on AI Labyrinth's** actual operation beyond Bright Data's internal observation of no degradation. - The demo is the vendor's own, presenting a **commercial interest** in the failure narrative he describes — the claims and the product pitch are intertwined throughout, as he presented them. # Speaker Attribution & Provenance Record ## Video - **Title:** Your Agent's Biggest Lie: "I Searched the Web" — Rafael Levi, Bright Data - **Video ID:** `01a15ca9-09cc-783c-a84e-928443e1ed88` ## Speakers & Roles ### Rafael (main speaker) - **Role:** Presenter / company representative - **Stated affiliation:** "I represent Bright Data" — described as a "web access platform to help agents or anybody collect public data on scale" - **Function:** Delivers the talk, runs the live code demo, answers audience questions, promotes Bright Data products ### Audience members (unnamed) - **Role:** Audience questioners - Three question exchanges, marked by `>>` in transcript: - Question about whether Bright Data runs social media accounts - Question about how Cloudflare Labyrinth detection works - Question about loading all MCP tools / tool count, and clarification about the experiment setup - Question about the free tier (5,000 requests per day vs. per month) ## Promotional Segments (flagged) - **Bright Data MCP** — core product demo: claimed **66 tools** (speaker also says **69 tools** in a later Q&A exchange; both figures appear in transcript) - **Free tier promotion:** "our MCP has a free tier of 5,000 requests" — clarified in Q&A as **per month** - **Pay-as-you-go** pricing mentioned - **QR code** shown on screen for sign-up - **GitHub page:** "GitHub brightdata" - **Skills page** promoted; upcoming second session at ~1:00 teased (Walmart scraper demo) - **LinkedIn** contact invitation: "you can connect with me on LinkedIn" ## Referenced Sources & Third-Party Claims (unverified, speaker's claims) - **Cloudflare:** "blocks AI crawling for about 20% of the web"; released an "AI labyrinth" that traps bots and feeds fake data - **ChatGPT citations:** "60% of citations on ChatGPT is not working" (speaker's figure) - **CAPTCHAs:** first appeared "a decade ago" - **Websites used in demo:** rightmove.co.uk, LinkedIn, Instagram, Amazon, TikTok - **Products/LLMs mentioned:** GPT-5 (demo model), Google/Bing/DuckDuckGo search (via MCP tools) - **Legal context:** lawsuits claimed — "LinkedIn suing these people... Amazon suing" ## Provenance & Availability Notes - **No curated metadata supplied** with this video record - **Comment availability:** not stated in source material - **Live demo failure noted:** presenter's ChatGPT comparison step "looks like it's stuck for some reason" - Speaker states Bright Data collects **only public data**, not data behind login, which it "considers illegal" - Figures cited (20%, 60%, tool counts 66 vs. 69) are the speaker's claims without cited sources Okay, so let's just work with it like this. It's a small room. Hi everybody. Welcome. My name is Rafael. I represent Bright Data. Bright Data is basically web access platform to help agents or anybody collect public data on scale. And I'm here to talk about the LLMs misleading people and all the time convincing them, "Hey, I did a search. Hey, I did this." While it didn't. Why? Because LLMs are programmed to please people, please users. So they make enough things and this is the biggest issue right now I'm seeing with LLMs. And I'm building applications all the time. I would rather LLM tell me, "No, I can't." But it never does. It always tries to make things up. So currently the web is actually fighting the robots and automations. And it's been something that going on for years. Everybody knows CAPTCHAs. The first CAPTCHAs showed up year 10 like, you know, decade ago. And it just keeps growing. And now we have AI blocking AI and there's a whole world going on. And the web access is actually not as simple as it looks. So they're getting CAPTCHAs and they don't actually report CAPTCHAs. So it tries to go and find the data different way. Sometimes it goes into training data. And this is where the worst thing is. When it uses training data and tells you that this is the current situation. Training data is from 2024. We're in 2026 and it doesn't add up, right? So these are some of the new things that I was just literally checking out, right? So Cloudflare blocks AI crawling for about 20% of the web, right? So 20% of the web is literally not accessible by AI by default fetch that's built into it. Cloudflare also right now released an AI labyrinth to actually trap bots and mislead them and provide to them fake data. So, then your results are getting even worse. Okay? So, the invisible failure group, right? There's no error, no warning, just the wrong answer, right? So, the agent sends a request, it gets a capture, uh even an empty page, it doesn't tell you, "Hey, I just got an empty page." It will try to make up something, and this is where the most of hallucinations come from. The need to please and lack of data. So, it literally makes things up. I've seen it literally make numbers up, provide fake citations. You click on the citation, it's a 404, the page doesn't exist, and I'm sure all of you have uh I've seen that happening recently. I mean, literally like 60% of citations on ChatGPT is not working. How many of you have tried to purchase a product? Hey, find me this product online, I want to buy it, and give me a link to the product. You click on the link to the product, and there's no product. Like, so what is this product that you're talking about? The URL doesn't exist, the product doesn't exist. So, where do I buy this product for the 50 bucks? Doesn't exist. Um so, what I want to show you, and I'm going to show you in code. How many of you are familiar with coding like um VS code? No? Okay, perfect. So, nobody's going to get lost if I'm going to switch to that. So, I'm going to do a demo basically with MCP and without the Bright Data MCP, and we're going to compare what is going on there. So, first I'm going to I I I want to show you that I have exactly identical prompts for both of the uh scripts, right? So, without MCP and with MCP. So, the the I I give it five tasks. Uh Property, rightmove.co. Let's go check out some of the properties, LinkedIn, Instagram, Amazon, and TikTok. These are the five basic uh sites that I wanted to access. They are very heavy on anti-bot systems, and I want to show you the difference. So, first I'm going to run it without an MCP, And I'm going to literally to let the AI talk for itself. And GPT-5 is a bit slow, so it's going to take some time. But basically, what we're trying to do is we're trying to access this URL. It's some local properties that I did literally half an hour ago. Um LinkedIn company in Israel, let's check it out. Instagram account, uh some Amazon product, and some TikTok. This is not limited to just these five websites. It's just something that I picked that is I know for sure will not work without MCP. So, as you can see, without MCP, I don't have live web data access. Uh doesn't have any browsing tools, right? So, this is with tools not available just by default out of the box uh uh GPT-5. I mean, it's a strong but it's a strong LLM. And zero success, five failed. Same exact thing, exactly the same prompt, I'm running with our MCP. Uh our MCP has 66 tools. While it's running, I just want to go off some of the tools that it has. Search engine. Search engine is basically the LLM is able to do Google search, Bing search, DuckDuckGo search. Real searches, not just like, you know, search the the web that it's doing in the background. Um it has It also has scrape as a markdown. That's a very strong one. Basically, it can send a curl to any URL and get just the markdown without HTML tags. So, you're not wasting tokens on parsing the HTML. Uh search engine batch. If you want to do, like, you know, like 100 keywords, you can literally can literally send 100 keywords and get 100 keyword batch results. Like, so the scaling is also huge. Discover. It also has pre-built APIs for many websites as you can see. And of course, it has a scraping browser infrastructure. So, it's a remote browser that your LLM can open and navigate. The remote browser solves capture by itself. It has a unique fingerprint. So, it can open 100 browsers, navigate the same website without getting blocked. So, the whole idea is that with our MCP, not only can it do a single sessions, it can do multiple sessions in parallel. And as you can see, let's see. We have success for the right move. We have success for the LinkedIn Misral. And Instagram also worked. Um Amazon product, it has the information for the product itself. And and then what I did is on the second part, I asked LLMs to compare the results from no MCP with MCP. So, that you don't take my word for it. Let's see what the ChatGPT will actually say. If it didn't get stuck. It looks like it's stuck for some reason. \>\> May I ask a question? \>\> Of course, please. I I love questions. \>\> For social profiles, do you have like uh accounts that you run? \>\> No, we only work with public data. \>\> Public data. \>\> Only publicly available data. Um collecting data behind login is not really legal. Why? Because when you sign up and create an account, you accept terms and conditions. When you accept terms and conditions, you need to really check it if it says, "Can you scrape? Can you Do you allow Do they allow robots to access the website?" And that's why you Maybe some of you heard there's a lot of lawsuits going on. LinkedIn suing these people. Everybody's suing them. Amazon suing, yes. \>\> But I think even um for example, LinkedIn and Instagram, I think they don't even show you really any public data even if you're on the \>\> Well, there's plenty of public data for for LinkedIn, of course. If you take for example, I will take this URL. I might get blocked. I don't know how's the local IP, but um and I open an incognito window, right? There you go. So, this is a public data that can be collected. Company, it's the same thing. It's the same thing. The only thing is they're very critical, right? So, if you are using, let's say, a Wi-Fi IP of a big event like this, it probably will block you because it's a data center IP, it's low quality IP. From home, you can probably access maybe 5, 10 profiles, but eventually it will also ask you to log in. Uh so, we only deal with public data. So, when you guys are using us, you are safe in the sense of nobody's going to come knocking on your doors to sue you, if that makes sense. No. We Again, we don't deal with data behind login. We consider it to be illegal, so we don't deal with accepting terms and conditions and data behind login. Only publicly available. We have a whole data set. So, if you guys don't want to do live data and you don't care if it's a few months old, we have data sets literally that you can just filter by, let's say, that you're looking If we're talking about LinkedIn people, you're looking for AI engineers in a certain area, you can filter it and just get the data set right away. So, and and your agent will actually have access to that. So, your agent can filter the data set and get you the data if you want to, right? So, it has access to all the tools. So, here's basically head-to-head comparison by the LLM itself. Without MCP failed, no live web access. With if listed, failed, successful, failed, successful. So, that's basically it. Uh antibot bypass, CAPTCHA solving, right? So, again, our system automatically solves CAPTCHA, so if your bot navigates to a website with a browser and has a CAPTCHA, our our browser has built-in capture solving solution so it will automatically solve the capture and your bot can continue on browsing without getting blocked. How much time I got? I don't know. Um So just to kind of summarize it, right? This is the biggest hallucinations that you guys see. The agent gets blocked, it needs to please you and it makes things up. And um fake content also, right? So now uh if you want to Google what is Cloudflare Cloudflare AI Labyrinth it's basically a system once it detects a bot it's not doesn't block it. It literally feeds it fake data. So bigger hallucinations. Right? And um the easiest fix for this is just to make sure that your agent doesn't get blocked. And it's as easy as to implement our MCP. Uh our MCP has a free tier of 5,000 requests. So if you guys want to try it out, um this is you can connect with me on LinkedIn if you want to. Or is this the Hold on a second. Is this the sign up for the MCP? I'm mostly in these QR codes. One second. Yeah, internet. Yes, please. \>\> Uh how does it detect if if there's Cloudflare Labyrinth or not? \>\> Um so the way we approach it is that we make your agent look like a human being. Literally like uh there's mouse movement pre-recorded, there's typing when it types it's like it's uh mimics a real human behavior. So uh the Cloudflare literally just doesn't even ask are you a robot or not, right? So this is our our approach. Instead of trying to understand how they detect we make the agent look as human as possible so that it doesn't trigger the actual blockage. Misleading data is one of the toughest things that you can actually encounter. A lot of websites right now in Asia is doing that, right? Hotels, they're literally providing different prices. You go check out on your phone, you get one price. You can check from your computer, you get a different price. You can ask for a proxy, you get a third price. Which one is correct? It's really hard to tell. When it comes into the domain of misleading, uh the best bet is to make sure that you your agent looks like a human, and then hope for the best. That That's basically what the approach is right now. Um AI levers was literally released a few like a month ago. Uh I don't have much of statistics on what is like, you know, exactly how it's working. All I know is that it didn't really affect us. We don't see any kind of change in data, and we're collecting petabytes of data on a daily like We have so many customers always scraping. We're caching data, so we're always comparing the results. We don't see any degradation in the results. Uh so, I think we're doing a good job in that case. Uh this is a QR code that you guys can sign up for. Uh we have a GitHub page as well. Uh GitHub brightdata.com. Oh, just GitHub brightdata. And another thing that I would recommend for you guys to check out is the skills page, right? So, what we did is we created skills. And I'm going to have another session in a couple of hours. If you're interested in seeing what the skills does, is basically you can take any agent, tell it to go here, and this page will teach it on how to build a scraper, or how to build a pipeline that will collect you the data. It has all the information it needs, all the APIs. And uh if you will come back to my next session, which is at 1:00, I think. Something like that. I'm going to literally demonstrate how it builds the pipeline. Literally in front of you, I'm going to tell it, "Hey, listen, let's build a Walmart collector for ABC." And it's going to build it, and it's going to scrape it. And instead of parsing each individual HTML, it's going to build a parser, and it saves about 99% of the tokens. Because I see a lot of people is like, "Hey, I need to parse 10,000 pages, but it's so token heavy." Don't parse with LLM. LLM builds a parser and then the script runs it. But that's the next session. Any questions? \>\> Uh question on the performance. Uh I saw that your MCB exposes 69 tools, which means if I if I need to do the same surgical ability for my for my agent do I need to load all the 69 tools \>\> of course filter it. You I just showed 69 tools because just to show it. Uh if I need just to scrape markdown and search, I would just literally load two tools. Uh otherwise you're flooding contacts with irrelevant data, of course not. I'm sorry, I didn't hear the beginning again. \>\> Uh the the experiment beginning \>\> Yes. Um so we didn't do any searches. Um I literally told it, "Hey, go to this URL, see if you can load it." Right? So that I didn't use the search uh in this demo. \>\> Sure. \>\> Uh but of course again even with our MCB they can actually do a Google search. \>\> Yeah. \>\> And that's the one of the biggest benefits is because we're used to Google results. So by default when you're asking LLM, you expect it to do a Google search, but it doesn't. So the results with the MCB much much better and I I recommend sign up, try it out, it's free. See the results, compare what you guys get. \>\> Is that for 5,000 requests per day or \>\> Per month. Per month. Which is, you know, like for an MVP for a little experiment is more than enough. Uh we also have pay as you go, so if you do need a little more it's it's nothing like, you know, it's Yeah, yes, but for a is perfect. I always I do a lot of hackathons and I'm always recommending, "Hey, listen, set up an MVP and tell your agent to go build whatever you need. It does a much better job than without."