MCP Server Implementation

Morning

Back on March 24th, I had asked the following question:

I was just thinking with all the talk about MCP servers recently, is this something we should be able to hook up to ?

Since then, it has become a reality, so have been spending some time getting MCP Servers going - having fun.

Kind of obsessed with trying to make the MCP Servers run within the Pickaxe Environment.

So far, have Make.com, Relevance AI, Dumpling AI and Apify ecosystems up and running along with some standalone MCP Servers such as Google Maps, AccuWeather etc.

Biggest challenge with the MCPs is that you can duplicate them. Hopefully that can be fixed easily.

If anyone is interested in doing an MCP Integration session, please let me know. If there is enough interest, I can setup a webinar in the near future.

4 Likes

I get MCP, but I also don’t.

I don’t get what MCP usecases I can envision (can you suggest what example usecases you have in mind?)

I don’t get why the paradigm of MCP vs API is different - in that if I wanted to connect to Youtube analytics in a web-app I go to YT API which is provided by YT. But it sounds like if I want to connect my bot to Youtube, I have to create my own MCP server/client that connects to YouTube and you have to do it yourself and everyone else has to do it themselves - why is that more efficient? Why can’t YT provide a MCP client directly…

Or maybe I don’t get MCP at all.

Please enlighten me if you can spare some time… :slight_smile:

Hi @sumergoconicio

Thanks for question.

I am no MCP expert by any means and not really into all the details about how it works. I am more ok, I can do this with it or not and go straight to use cases and implementation.

When MCP first came out, I thought quite neat but it looks like you can only use it for yourself - if I was using it in cursor etc. and saw no real advantage in terms of how would build this into something that others could use and you could monetize.

As I mentioned in my post, I asked the question back in March as well as brought it up in one of the office hours sessions. So was glad to see that it is now available within Pickaxe.

Only annoying thing is if I had know about MCPs being an action, I would have purchased some additional actions on the last go around as I only have 3 per Pickaxe - all actions, so am a little restricted. 10 would be a good safe number to have.

So before the MCP option was available, if I wanted to send information to say a google sheet, I could connect it to a Make.com Scenario and have Make do the heavy lifting in the background. Then they introduced agents so that you could build out different scenarios for each agent to work on.

With the MCPs, they typically come with a number of tools / calls that you can use, so in the use case of connecting to a google sheet, you can now call the google sheet MCP where you can now send data to a google sheet, retrieve data, ask it questions like what is in Cell A2 of Sheet1 of Google Sheet “Contacts” etc.

Another example with Relevance AI - I could never get it to connect with the API, but now they have an MCP for each tool that you build. With the MCP, you can add in multiple tools to the same MCP, so for me, it makes thing simpler for sure.

Make.com now also have an MCP that you can run at the Org, Team or Scenario level, so that works well too. So you can build out your Make.com scenarios, hook up one action and now you can access all the scenarios you want to use by making a specific call - telling the Pickaxe what tool to use and for what use.

While typing this up, I connected the ElevenLabs MCP to a Pickaxe which allowed me to do the following:

  1. Text to Speech: Convert text into audio using different voice options.
  2. Speech to Text: Transcribe audio files into text.
  3. Sound Effects: Generate sound effects based on text descriptions.
  4. Voice Search: Search for existing voices in the ElevenLabs library.
  5. Voice Cloning: Create an instant voice clone using audio files.
  6. Audio Isolation: Isolate audio from a file.
  7. Conversational AI Agents: Create and configure custom conversational AI agents.
  8. Knowledge Base Management: Add knowledge bases to agents for improved responses.
  9. Phone Call Integration: Make outbound calls via Twilio using ElevenLabs agents.
  10. Voice Previews: Generate voice previews from text prompts.
  11. Add Generated Voices: Add generated voices to your voice library.
  12. List All Agents: Retrieve details about conversational AI agents.

I am still playing with it myself and finding out new ways to make things work together while keeping you in the Pickaxe environment.

Next step will be to try and build out my own MCP server - but have no idea what to call etc.

I hope this has helped answer yor question.

1 Like

@sumergoconicio great answer by @ihmunro.

MCP is often described as the USB-C port for AI applications—essentially, it’s all about standardization.

Take the ElevenLabs example: to access 12 different API endpoints, you’d typically need to create 12 separate actions and write code for each API call. In a no-code environment like Make, this would mean building 12 individual scenarios.

With MCPs, however, you only need to connect to a remote MCP server. The application—like ElevenLabs—exposes all its APIs through this server, eliminating the need to manage individual API calls. This drastically reduces development and deployment time.

Another key benefit is that you don’t need to deal with the specific requirements of each API, which often vary. Instead, you simply connect using a standardized protocol—this is where the value of standardization really shines.

To connect to a remote MCP server, you do need an MCP client. Traditionally, only those with access to a local machine (like Claude Desktop) could connect.

Pickaxe changes that. It’s one of the first cloud-based MCP clients, enabling you to build applications that connect to remote MCP servers—and making those applications accessible to users without relying on local infrastructure.

5 Likes

A few of us were just talking about this yesterday. I would love to attend an MCP integration session. This sounds amazing. As was said below, I get it and I don’t. For me it is seeing where all the puzzle pieces fit. I get the big idea, but not the implementation and a session to see that in action would help so much.

1 Like

Thanks for the additional info @ab2308 - appreciate it.

Appreciate the comments @jeanette

Lets see if there is some more interest before I set a date.

If it is just 2 or 3 people that is ok too.

I’d be interested to learn more about MCP server implementation.

@user14 Excellent - any specific area of MCP Implementation?

Generally just want to learn more about proper implementation. I’m specifically trying to allow our pickaxes access to internal data sources such as Airtable, Freshdesk, and similar. Maybe some basics and then some best practices.

@user14

Good new is that Freshdesk has an MCP - I will see if I can connect it to it tonight when I get home.

Airtable does not have an MCP - however there are some homegrown out there, so just need to find a good verified one.

Would love to see how you are using Make, I’m building a series of bots for client customization and we use a lot of make scenarios. Would be amazing to let them handle the front end. I’m sure it will spark some new ideas.

1 Like

@user14

I had some time so got the Freshdesk Connected - so I know the code is working.

However, I only just signed up for a free account and I have not option for getting an API, so cannot try any of the calls for it.

Hi @bryanmvp

I have the Make.com MCP working pretty good.

This will be one that I cover.

1 Like

Yeah, for airtable I have been using this homegrown one: GitHub - domdomegg/airtable-mcp-server: 🗂️🤖 Airtable Model Context Protocol Server, for allowing AI systems to interact with your Airtable bases

Is there an official Freshdesk MCP? How to find this?

Yes

Hi @jeanette

Is there anything specific you wanted to connect to or just more curious about what can be done with MCPs ?

I am very curious how to set up my own MCPs. For example, we are currently working a project that involves eleven labs, google storage, to layer music on longer audio outputs. We are having to break up the audio to make it work. Can this all be put into one MCP? I feel like I understand how to use other MCPs if given to me, but not how to find them or how to build them. Does that make sense?

1 Like

I’m in on this one! What can be done? Especially with Make. :heart:

1 Like