We built an MCP for our scheduling SaaS and are now pivoting our business model
I wanted to share something exciting we've been working on at TimeTime (our scheduling SaaS platform) that completely changed our perspective on where we should be heading as a company.
What is MCP?
For those who aren't familiar, the Model Context Protocol (MCP) is a groundbreaking open standard introduced by Anthropic that's essentially "magic" for LLMs. It enables AI models like Claude to directly interact with external systems without human intervention.
The key breakthrough: MCP eliminates the need for humans to wrestle with documentation and APIs. We can now delegate these tasks to the AI through an MCP, allowing the AI to:
- Make API calls to our systems
- Obtain implementation examples
- Read and understand Swagger/OpenAPI specifications
- Take concrete actions based on its understanding
In our implementation, Claude can explore our TimeTime API independently, understand how it works, and build working implementations without us having to explain every detail. It's a complete game-changer for how we think about the relationship between humans, AI, and software systems.
You can learn more about MCP here:
What we built?
We created an MCP that enables Claude (through Cursor) to interact directly with TimeTime's API, allowing the AI to independently read our documentation, make test requests, understand our scheduling system's architecture, and build custom implementations without human intervention. The most incredible outcome? With our MCP integration, what used to take days or weeks of development work now happens in less time than it takes to drink a cup of coffee—we can deploy a fully functional, completely customized scheduling system in under 10 minutes. This revolutionary approach completely transforms the development process, making it exponentially faster and more efficient while maintaining high-quality, tailored implementations that meet specific business needs.
See it in action
I've recorded a couple of demos showing how this works in practice:
Demo 1: Asking the LLM to build a scheduling system
In this demo, you can see how we used Cursor with Claude to create a complete booking chatbot. From a simple prompt, the AI explores our API documentation, understands the requirements, and implements an almost functional solution in less than 5m
Demo 2: Watch the LLM fix errors like a human developer
This second part showcases how the LLM can troubleshoot issues just like a human developer would. Given a http error, the LLM explores our API, makes test requests, identifies errors in the implementation, and fixes them independently.
Reactions at JSConf
We presented this technology at JSConf Spain this past weekend, and the response was overwhelming. Developers immediately recognized the paradigm shift this represents for how software can be built and customized.
This got out of hand, and some developers ended up kicking our booth, saying that we're going to take their jobs.1
Check out the code
We're excited to announce that we've released our MCP code on GitHub for any company that wants to try it out. Getting started is incredibly simple - just add your Swagger/OpenAPI specification and you're ready to go!
Check out the full repository at: https://github.com/timetime-software/timetime-mcp
This repository allows any company to distribute their own MCP implementation, you just need to add a valid openapi.json file and a prompt explaining what your company does.
Some implementation notes
It's worth mentioning that the current reference implementations for MCP aren't fully complete yet. We encountered some challenges with bundling due to CommonJS and module import issues, we are more than happy to help you with that. Despite these growing pains, the potential of this technology is undeniable.
A new era of programming
What we've experienced with MCP represents a clear before-and-after moment in programming history. The ability for AI to directly interact with systems, understand APIs, and build implementations without constant human guidance fundamentally changes how software will be developed going forward.
I'd love to hear your thoughts on this direction! Has anyone else been experimenting with MCP or similar approaches to AI-driven development?
Our pivot
This success has led us to reconsider our entire business model. Offering UI components for scheduling no longer makes sense as our primary focus when we can provide something much more valuable.
We're now pivoting to position TimeTime as a scheduling infrastructure layer specifically designed for AI coding. Instead of being yet another scheduling tool, we'll be the platform that makes it incredibly easy for AI to implement scheduling functionality within any application.
TLDR: We built an MCP for our scheduling SaaS that lets Claude interact with our API directly. Now we can create customized scheduling systems in under 10 minutes. We're open-sourcing our implementation and pivoting our business to focus on being a scheduling infrastructure layer for AI coding.
Footnotes
-
JOKE ALERT! No developers were harmed in the making of this blog post, and our JSConf experience was entirely positive ❤️ ↩