5 Useful Things You Can Do with MCP Servers
Voice calls, Slack messages, web scraping, medical search, and even Minesweeper — practical examples of how MCP unlocks real-world functionality within LLMs.
I started writing about the Model Context Protocol (MCP) for a simple reason: I wanted to see if I could use it to do cool things.
As a reminder, MCP allows LLMs to connect to third-party tools and data, adding fresh context that helps the AI act more effectively on your behalf.
This post shares five (cool!) real-world examples of what that looks like in practice when using MCP servers and LLMs in tandem.
🔌 Zapier: Connect AI to (Almost) Anything
Automation platforms like Zapier or n8n are already powerful because they let you chain together actions across many apps. But with MCP, they become even more useful: now, you can access hundreds of tools directly from within your LLM.
Instead of configuring separate Slack, Jira, and Google Calendar MCP servers, you just connect your Zapier account, and and unlock access to many tools at once from within your LLM.
Here’s a video example. In it, Bryan Helmig from Zapier demos the Zapier MCP server inside Cursor, an AI-powered IDE (coding development environment), to:
Post a message to Slack
Create and edit a Jira ticket
Zapier connects to tons of other tools, so you can easily extend this to scheduling meetings in Google Calendar, sending Gmail follow-ups, updating Salesforce records — all via natural language.
Bonus: Here’s a great marketing example using n8n to build a dashboard from Google Analytics data.
Setup links: Zapier MCP, n8n MCP
🗣 ElevenLabs: Voice Calls & Speech Synthesis
Voice AI tools like ElevenLabs are powerful on their own, enabling capabilities like text-to-speech and voice cloning.
But if you plug ElevenLabs into an LLM using MCP, and you can do things like what’s shown in this video from ElevenLabs, like:
Create a custom ElevenLabs voice agent
Make an outbound call to order a pizza
Handle the entire conversation: speech-to-text → text processing → text-to-speech
This opens up valuable use cases, both for businesses and consumers. Imagine:
A sales rep provides the phone number and sales pitch to use for a cold call.
The ElevenLabs MCP creates a voice agent and calls a sales prospect
If they’re interested, Claude uses the Zapier MCP server to check the rep’s calendar and schedule a meeting
Or, as a consumer: your assistant could call the doctor’s office and check your availability to book an appointment. This future’s not far off.
Setup link: ElevenLabs MCP
💬 WhatsApp: Message, Search, Automate
MCP servers can text, too!
This demo from Luke Harries shows an unofficial MCP server for WhatsApp that lets you:
Search your contacts
Send messages
Read and search recent WhatsApp conversations
This isn’t just helpful for consumer use cases (no more flipping tabs to copy/paste from chats!).
It’s also powerful for businesses that use WhatsApp for:
Order tracking
Customer service
Shared communication logs
Sadly, there’s no equivalent MCP server for iMessage — Apple’s ecosystem remains (predictably) closed off.
Setup link: WhatsApp MCP (unofficial)
⚠️ Two important warnings about the WhatsApp MCP server:
Security: Researchers have shown how a malicious WhatsApp MCP server could exfiltrate your messages. This isn’t a known real-world attack, but proceed carefully — and only use trusted sources.
Privacy: Messages accessed by an MCP server may become visible to your LLM. Be cautious about sharing sensitive conversations.
🌐 Firecrawl: Scrape the Web for Context
Most AI tools offer web search — but if you want to pull in data from a specific webpage, a tool like the Firecrawl MCP server is likely a better choice.
For example, this video from Firecrawl co-founder Nicolas Camara shows how you can use the Firecrawl MCP server to:
Scrape a webpage
Feed the contents into Claude
Recreate the page using code
💡A personal example: This reminds me of a time when I wanted to get a list of the sponsors for a conference, but they were only available as a grid of linked logos on the company’s website. At that time, I went to the site, opened dev tools, and copy-pasted the relevant section of HTML into ChatGPT to ask for a CSV. MCP compresses that entire workflow into a single prompt.
In addition to web scrapers like Firecrawl, there’s also a similar category of MCP servers that can do something similar by using headless browsing to interact with more complicated websites (e.g. with login flows or JavaScript-heavy sites that are less conducive to scraping); Browserbase is a popular example of a company that does this.
Setup links: Firecrawl MCP, Browserbase MCP
🧬 BioMCP: Search Biomedical Databases in Context
While LLMs have broad training data and are great at reasoning, they’re not great at surfacing specialized research.
One way to do this is to have purpose-built tools for a given audience, like how OpenEvidence offers a ChatGPT-like interface, but with a focus on medical information.
However, with MCP servers like BioMCP, you can also search and pull in specific data sources that are relevant to your work. For example, BioMCP pulls in data and articles from:
PubMed
ClinicalTrials.gov
MyVariant.info.
This shows an example of what that might look like in action
Similar tools exist for other research sources like ArXiv as well, as do MCP tools to streamline citations.
Something like this is hugely useful for PhD students, researchers, or clinicians drafting papers or planning trials.
Instead of manually toggling between search engines and the LLM interface, and downloading and uploading reports, the data is pulled in and ready to use, and you can even generate citations.
🎉 Bonus: Fun + Niche MCPs
There are also lots of fun and creative (if less practical) MCPs out there.
A few of my favorites:
🎵 Spotify MCP: Play music or have the AI comment on your recent listening
🌤️ Weather MCP: Pull live weather data — great for trip planning prompts
🎮 Minesweeper MCP: Play Minesweeper through an AI assistant:
Because it’s surprisingly easy to create an MCP server yourself, you can also create an MCP server of your own for anything that interests you and has a public API, like analyzing bird vocalizations, or getting real-time soccer (football?) data.
My Own Setup (So Far)
I’m still experimenting with how MCP fits into my daily workflows.
So far I’ve been using:
My custom Canvas MCP server to summarize class due dates
The Filesystem MCP server to save outputs from Claude
The Zapier MCP server to experiment with day-to-day task orchestration
I’m constantly coming across new and interesting MCP servers to try, so I’m sure I’ll add to this list over time.
More soon!