MCP Tools Usage Guide

So, since the Model Context Protocol (MCP) was announced at the end of November 2024 I immediately saw the potential that this could offer me. Me being a person that likes to gather information to add to an LLMs context before doing anything. When Cline released v2.2.0 was when I started making my own tools! if you take a look at the tools on my GitHub profile you’ll notice a lot of them are for search in some form or another.

The reason for this? As stated before, is to load up an LLM with context on a task before taking action. For this post I’m not going to get stuck into the details on why it’s a good practice to do this, but, a great example of why you would want to do this is if you’re “vibe coding” something and you want to use Tailwind. Tailwind v4 was released at the end of January 2025, there’s no LLM that has any of that information in it’s knowledge right now. Also LLM training biases mean that LLMs will always favour what is already available to it.

This means a “make my SaaS using Tailwind” would result in Tailwind v4 being installed and then nothing would work because the LLM goes and implements Tailwind v3 config and therein lies the issue.

So, with a total of 7 months of working with and creating MCP tools, I’m now, according to the internet hype engine an expert! 😂

Ok, so, with that rather large pre-amble done, what I want to detail is what MCP tools you should use and why.

I can spin up a browser with Claude? What?

Yes, there’s MCP tools like Puppeteer and Playwright that you can install and navigate to a site to parse information from. The thing is when you just need to get that information into context there can be a bit of a wait, watching the browser spin up and then seeing it navigate around a site is novel at first but tedious each consecutive time you need to gather information.

From my point of view (and experience) the best way to get information from the web into your LLM of choice is using a search API provider.

While browser automation tools like Puppeteer or Playwright are powerful, they’re often not the best choice for simple information gathering. Here’s why:

• Resource intensive - spins up a sandboxed browser instance
• Slower iteration cycles - waiting for page loads and rendering
• Higher latency - needs to parse and process DOM elements
• Potential for hallucinations - especially with vague queries

Better alternatives for information gathering

Instead of browser-based tools, prefer these more efficient options:

1. Official MCP Fetch Tool

   • Optimized for text-based content
   • Returns LLM-friendly formats (text/markdown/json)
   • Faster response times
   • More reliable results

2. Search API Providers

   • No browser overhead
   • Direct access to indexed content
   • Structured response formats
   • Better query understanding

Consolidating search providers with mcp-omnisearch

Rather than running multiple MCP servers for different search providers, I created mcp-omnisearch to consolidate them all:

• Brave Search
• Perplexity
• Tavily
• Kagi (invite only)
• Jina AI
• Firecrawl

Benefits of this means:

1. Single Configuration

   • One config file
   • Simpler setup

2. Flexible API Usage

   • Use any provider you have keys for
   • Mix and match capabilities
   • Easy to add new providers via API keys

3. Resource Efficiency

   • Single process
   • Shared resources
   • Lower memory footprint

You don’t need to use it! The whole reason I made it was so that I wasn’t sharing my API keys with random code from the internet! Make your own or use a trusted source. I made this to scratch an itch.

Conclusion

LLMs will often try palm you off with their base knowledge or guess the information you’re looking for. Don’t stand for it! 😂 Make sure they gather the context needed.