Model Context Protocol enables a Large Language Model (LLM) to do a lot more than just answer questions. Acting as a translator between the model and the digital world, it can abstract data from a particular service by exposing “tools” that the LLM can use. Instead of manually pasting snippets from another application, the model can choose to access that information if required. SearXNG is one such MCP server that enables a model to browse the web, but Context7 is another, fantastic MCP server that enables your model to access up to date documentation.

Context7 is very simple: it offers up to date documentation for many different programming languages and services. If you’ve ever tried to use an LLM for writing code in a less-popular programming language or in one that rapidly changes and improves, then you may have had experiences where the advice you received or the code the LLM wrote simply didn’t work as it wasn’t valid. With Context7, your LLM can query documentation in an easily understood way, improving the chances of getting code that actually works, rather than old, or even hallucinated, responses.

How Context7 works

It’s just documentation in markdown

Credit:

Context7 is quite simple, as it all it does is offer a way for an LLM to query up-to-date, version specific documentation with code samples from the source, including them in the LLM’s context for deeper understanding of what it’s being asked to do. While it works great with a local LLM, it’s a fantastic tool when paired with any LLM that offers MCP support, such as Claude. You link it to Context7, tell it to “use context7” somewhere in your prompt, and that’s it.

For example, CounterStrikeSharp is a server-side modding language that you can use for Counter-Strike 2 for servers, and as a relatively new language, it can be hard to get the likes of ChatGPT, Claude, or a local model to actually write code correctly in that language. Context7’s documentation usually encompasses the entire language or toolkit in a searchable Markdown format, meaning that an LLM can help you out, as it now has the context to see the syntax of the language and how it works.

While Context7 is aimed at coding primarily, it can do more than that. It can search documentation in general, so you don’t just need to ask for code. You can debug existing code that doesn’t work, in order to help you fix whatever problem you’re dealing with.

For an example that doesn’t strictly involve programming, I used the Challengermode documentation to prove how powerful Context7 can be, as it’s a fairly niche platform that it has documented on the site. Challengermode is a popular gaming platform where users can link their Steam accounts (or other accounts) and compete against players in a game.

My query was simple, but one that I knew my local LLM (Magistral Small 2509) wouldn’t have. I asked for the base URL that the service uses for its API calls, and it got it completely right using Context7 to request that information beforehand. It’s basic, but it works, and it’s the exact kind of tool that can enable you to work better by using an LLM for gathering information about an API.

Using Context7 for actual programming

It really impressed me

I often see people say that they tried to use ChatGPT or another, similar LLM for working in Home Assistant or ESPHome but they struggled, and truth be told, I’ve never been able to really get it to work, either. I’ve found that the reason for this is that they tend to base themselves on old methods, or use inefficient ways to achieve something. Worse still, it can often straight up use methods and fields that literally don’t exist. I’ve been experimenting with Context7 and ESPHome, though, and it’s been fantastic for some very basic stuff that trumps every other “vanilla” LLM example I’ve seen.

As an example, I asked my local LLM to use Context7 to help me write an ESPHome YAML config to display information from my Pirate Weather sensor on Home Assistant, and design a layout to show it on the WT32-SC01 Plus using LVGL. I’ve found LLMs will use all kinds of crazy syntax as soon as I ask about LVGL, and I’ve yet to get a working configuration in typical usage, yet even with a local LLM, it did a ridiculously good job.

Of course, I did notice that it got “lazy” in one part. When mapping the icons to weather statuses, it dropped the following block of code in the middle:

# Weather condition icon (centered at the top) - obj: type: label id: weather_icon text: "?" align: CENTER x: 10% # Adjust as needed y: 5% width: 80% height: 30% text_font: weather_icons lambda: |- if (id(current_weather_condition).state == "clear-night") { return "U000F0594"; } else if (id(current_weather_condition).state == "cloudy") { return "U000F0590"; } // Add more conditions as needed return "U000F0594"; // Default to clear-night

It just didn’t add the other weather conditions, leaving it to the user to complete those. It’s not a big deal at all, but overall, it did a pretty good job at writing code considering that this is a local model almost nailing it when I’ve seen the likes of ChatGPT struggle. This is almost certainly a closer solution than any LLM would get without Context7, so you can imagine what the likes of Claude can do when using the desktop app and linking it up yourself.

Context7 is completely free, and while the free edition has usage limits, I haven’t run into them even while testing it out and running many different samples through it for the purposes of this article. It’s worth checking out, and if you want to use a local LLM to help you with coding, I’d argue that it’s a must. I still don’t think it supplements actual programming experience and knowledge, but in many ways, it’s the first time I’ve seen an LLM be capable of being an actually good coding assistant in these niche use cases.