๐ ๐๐ฃ ๐๐๐ป'๐ ๐ ๐ฎ๐ด๐ถ๐ฐ, ๐๐'๐ ๐๐๐๐ ๐ ๐ฅ๐ฒ๐ฎ๐น๐น๐ ๐๐ผ๐ผ๐ฑ ๐๐ผ๐ผ๐ฟ
I started with n8n and became obsessed with AI pipelines.
I did not want to be a user. I wanted to know how they work. This led me to the Model Context Protocol (MCP).
Think of MCP as the USB-C for AI. It is an open standard to connect AI applications to external systems.
It allows an LLM to talk to your databases, Figma, GitHub, or PostHog. It turns a chat interface into a tool that acts on your behalf.
When building an MCP server, you focus on three things:
- Resources
- Tools
- Prompts
I started with Tools. These are functions your LLM uses.
You can build servers using two methods:
- STDIO: Best for local, fast tools for a single user.
- HTTP: Best for web applications and distributed systems.
I expected a difficult learning curve. I thought I would need heavy engineering knowledge. I was wrong.
Building a tool is simple. You follow a well-defined contract. You register a tool, define its description, and write the function. The rest is just boilerplate.
I tested this by building Gitstoria.
Gitstoria helps you attach reasoning and notes to your git commits.
- You make a commit.
- A hook records the hash in a local database.
- You ask Claude to log your work.
- Claude uses the MCP tool to read the diff and write a session log.
MCP is the door that gives an LLM access to your system. It can be a simple clock or a tool that manipulates entire web applications.
You have two paths in AI:
- Workflow automation: Use tools to reduce friction and gain efficiency.
- AI engineering: Learn how the models work under the hood.
I chose to go deeper.
Full post: https://dev.to/marcochavezco/mcp-isnt-magic-its-just-a-really-good-door-53gb
GitHub Full Code: https://github.com/marcochavezco/gitstoria
Optional learning community: https://t.me/GyaanSetuAi