Claude Code Is MISSING One Key Thing. Here's How to Fix It!

Author: Universe of AI · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Claude Code Is MISSING One Key Thing. Here’s How to Fix It! · Channel: Universe of AI · 09:48 · Watch on YouTube

Topics: AI agents, Task and workflow automation, Claude Code, RAG, search and information retrieval

Summary

The video presents context.dev as a way to help AI agents retrieve and use real web data, addressing a limitation the presenter identifies in agent workflows. It demonstrates connecting context.dev to Claude Code using an agent setup prompt and an API key, then testing the connection. With instructions in a project file and a simple leads CSV, Claude Code uses context.dev tools to research six leads and produce an enriched CSV with company, person, pricing, careers, and recent-news information, along with sources. The presenter highlights that missing information is marked “not found” rather than guessed, and mentions other possible sales workflows and context.dev’s page-monitoring feature.

Key points

The video frames access to reliable web data as a challenge for AI agents doing tasks such as research and lead generation.
context.dev is presented as a web data API for AI agents, with tools for scraping, search, crawling, monitoring, company data, and people data.
The demo focuses on four tools: scrape anything, answers, brand enrichment, and people enrichment.
The presenter connects context.dev to Claude Code by following an agent quick-start prompt, creating an API key, and running an MCP test.
A leads.csv file contains only names and company websites; a cloud.md file supplies the research instructions.
The instructions ask the agent to research each company and person, review pricing and careers pages, find recent news, score the lead, and draft a cold-email opener.
Claude Code creates an enriched CSV containing fields such as industry, location, role, match score, pricing, open roles, recent news, lead score, and sources.
The presenter emphasizes that the agent should mark unavailable information as “not found” rather than inventing details.
Suggested uses include evaluating new sign-ups, preparing for sales calls, and monitoring a competitor’s pricing page.

Chapters

Start	Chapter	Summary

Quotes

It’s a web data API built for AI agents. — 01:01

What comes back is a structured answer plus the source URLs it used. — 02:03

If the agent can’t find something, it writes not found. — 05:08

you don’t want the AI agent to create something that doesn’t exist. — 08:47

Concepts and entities

context.dev, Claude Code, MCP, CLAUDE.md, API key, Scrape Anything, Answers, People Enrich, Brand

Full transcript (original language)

00:00 In 2026, the rate of AI development has improved a lot. We probably have the strongest models right now that are quite capable. AI agents that can get things done very quickly, and even platforms like Cloud Code have gotten much more capable than before. But the real problem right now is that you can set up your workflow, have all the pipeline and the infrastructure ready to go. But when you ask your AI agent to do actual work, like retrieve data from actual websites, research people, or things like that, or build a lead

00:30 pipeline, it fails because it’s not able to pull that data. [music] So, how do we solve exactly that? Well, that is what today’s video is about, so let’s get into it. So, today’s video is sponsored by context.dev, and they built something for exactly the problem I mentioned in the intro of this video. By the end, I’m going to show you an agent running in Cloud Code that takes a plain list of names and turns each one of them into a research sales lead. Obviously, this is just one example. You can do a lot with this platform, but this is just so you guys have an understanding how to use

01:01 platform, how to set it up, and how capable this new platform is. So, what exactly is context.dev? It’s a web data API built for AI agents. Instead of your agents trying to figure out raw web pages on its own, it asks context.dev, and it gets back clean data that it can actually work with. There are a bunch of products under one API key: scraping, search, crawling, monitoring, company data. Before the matter for today, so I’ll stick to those. So, what you’re looking at right now is my context.dev dashboard, and you can see all the data

01:32 APIs are available over here. There are a bunch of options for web extraction, brand intelligence, people and news. For example, in web extraction, we have scrape anything, map URLs, crawl website, search web answers. Brand intelligence, we have retrieve the brand, search brands, extract style guide if you need to do that. And then we also have a people and news, enrich person, and search news. In today’s demo, I’m going to be showing you guys four of these. The first one is scrape anything. You give it a URL and it gives you the page back as a clean markdown, a

02:03 screenshot, or structured fields. Then there’s answers. You ask it a research question, you give it a JSON shape for the answer, and it goes in and searches the web for you. What comes back is a structured answer plus the source URLs it used. So, it works like a deep research tool except it’s one API call and you can check where every fact came from. Then there are two enrichment ones. Brand takes a domain or a work email and it gives you the company profile, name, description, industry, location, logo, social links. And then

02:35 the people one takes a name or an email and tries to find the actual person. It comes back with their role and background plus a match score from 0 to 100 that tells you how confident it is. If you put those four together and your agent can answer, “Who is this person and what does their company do?” without you opening a single browser tab, which is exactly what we’re going to build. So, how do we actually set up our agent to use context.dev? Luckily for us, context.dev have made it pretty simple. There’s an agent quick start and the

03:05 setup prompt is pretty much here. What this prompt does is it tells the agent to kind of guide it to this doc page and basically everything at the bottom is actually instructions for the agent. So, it’s a simple step. Copy this prompt, then open up Cloud Code. For example, I have Cloud Code open over here. Then I’m going to take that prompt, just paste it in over here. And the first thing you’re going to have to do is set up your API key. So, if you haven’t already, you can do that. I’ll just use the dashboard method, so I’ll open up the dashboard on context.dev, get the API key, and then send it in. So, I’m going to go to my

03:36 API keys over here and then just create a new one. And you can see I have one over here. I have disabled it, but you can just create an API key. Click on that. Name it whatever you want to. So, today I’m going to doing a demo, so I’m going to name this a demo key. So, now just to make sure that my session is actually connected to, you know, context.dev. I started a new session in that same folder and I’m going to paste this in, which is run one context.dev MCP test. It’s just going to, you know, just make sure that everything is working correctly. The tools it needs to run context.devs are enabled and

04:07 installed. So, it’s just doing that test. So, it looks like we are good to go. That means everything is connected pretty easily and that is why using that simple agent prompt sometimes makes sense. You can take that prompt, put it into codex use that, cloud code, whatever the case is, but just run a test after just to make sure that everything is connected properly. All right, now that everything is set up, I’m going to be working in cloud code and I’m going to show you the demo. So, I got a folder with two files in it and the folder name is lead agent demo. And in there, the first one is a leads.csv

04:38 file. This is usually all a sales rep starts with, a first name, a last name and a company website and that’s pretty much the whole file. I’m using well-known founders so you can follow along with this, but you can change this based off of your needs. The second file is a cloud.md file. Cloud code reads the file automatically before it starts working in a folder. So, instead of typing a giant prompt every single time, all the instructions live here. For every lead, find a company, then find the person and read their pricing and careers page, look up recent news, then

05:08 score the lead and write a first line for a cold email. There’s also one rule in here I care about a lot. If the agent can’t find something, it writes not found. So, it’s not going to guess or fill it in with incorrect information. Because the instructions are already in the file, I can just put in a simple prompt, which is research the leads in leads.csv and save the results to enrich leads.csv, which is going to be the output file that cloud is going to create for us. One thing to note is that all of this is going to be possible because we’re going to be using context.dev’s MCP tools. So,

05:40 now let’s put in our prompt and test this out. So, I put in my prompt at the bottom over here. I’m going to press enter. All right, let’s watch the leads closely because this is where you can actually see each tool from context.dev doing its job. So, if I take a look at this one, you can see the body that I gave it was domain, and then the name or the website is a linear app. You can see the description it was able to get is all of this over here. Linear is a product development platform design XYZ. At the bottom, we can also see that it was able to pull in some of the brand data about linear app over here. And at

06:12 the bottom, you can see it’s doing the same thing for Figma. And all of this is coming from our leads.csv file. Then at the bottom over here, it’s using another tool called people enrich. And you can see over here that we gave context.dev a simple, you know, first name, last name, as well the company they work for. And context.dev was able to pull all the information that you see at the bottom, like location, San Francisco, country, United States, and it’s using LinkedIn actually to get in that person’s information, as well its title. Who is this person? The CEO, as well current role, experience. All of this is being

06:45 pulled because of context.dev’s people enrich tool that’s being active at the moment in Claude code. Then it’s using the answers tool to basically pull in the specific information. So, I gave it this question in my instructions. What has this company announced or launched in the last 90 days? And what is the main product focus? And then gave it the current date, so only include items from roughly the 90-day period. And all of this is being pulled from the website. So, you can see I gave it a simple thing and it’s going to pull that specific information based off the constraint I have given it. And this is using the

07:16 answers tool from context.dev. Then it’s also using the web scrape feature from context.dev to find any job openings basically for all the websites that we have listed. All right, Claude code is done and it wrote all six leads to a new file that you see at the top over here, enriched leads.csv. But before we look at the file, let’s just look at the original file. You can see it has a first name, a last name, and a company website, and that’s all I had to work with. And this is the updated file. Each lead now has a company’s industry and

07:46 location, the person’s role, a match score, pricing, whether there’s an enterprise plan, open roles, recent news, a lead score, a short reason for that score, and a first line for a cold email over here. Plus a column with every source that I use at the right over here. The sources you can see over here. If you look at Linear, for example, from what it found, Khari is CEO and co-founder with a 98% match score. Pricing is also starting at $10 per user per month. There’s an enterprise plan and 23 open roles. It

08:17 also picked up three recent launches, Loops, Coding Sessions, and Party Inbox, and wrote an opener based on those. Each of those columns came from a different context.dev tool. Brand handled the company profile, People found the person and gave the match score, Scrape Anything tool read the pricing and careers pages and pulled out just the fields I asked for, and Answers handled the recent news with links back to where it found it. Claude’s part was just putting it all together and writing the email openers from the data that context.dev pulled. It also didn’t fill

08:47 in any gaps. If there was any information that was missing, it came back with not found, which is something that you want because you don’t want the AI agent to create something that doesn’t exist. And obviously I used a simple example here, but the same setup could probably work for a few other things. If you get sign-ups with work emails, you could run these steps on each new sign-up and see which ones look like a good fit for your sales team. Or before a sales call, you could run it on just that one company. Instead of reading through their website and news yourself, you would get a quick summary of what they sell, what they launched

09:17 lately, and who you’re talking to with links to check. And context.dev also has monitors, which watch a page on a schedule and send you an update when it changes. So, you could keep an eye out on a competitor’s pricing page without checking it yourself. But that’s a quick intro to context.dev. If you want to try this out for yourself, the link is in the description. There’s also a free tier with 1,000 credits a month and no credit card is required for that. So, you can try this out for yourself before you decide to purchase a subscription. And once again, big thanks to context.dev for sponsoring today’s video.

09:48 But, that’s it for today’s video. Make sure you guys are subscribed to the channel. Follow our new newsletter as well at universeofai.beehive.com. As well, subscribe to the main channel World of AI and support us on X by following the universeofai.z as well. Until then, I’ll see you guys in the next video.