Labbing an MCP
April 28, 2026
Answer the following questions in this reflection:
What is the value of an MCP server? Why can't an LLM do what it does?
Model context protocol servers are just a way to store tools for agents to use and access to databases, another autonomizing tool.
What is meant by the term model context protocol? Why are those three words in the name?
Agent model, context of the agent, and protocol for how to use the tools.
Describe two actions that you want an MCP server to offer that an LLM can't already perform. Each action should rely on at least one parameter. Do not choose weather lookup. Yawn.
I wrote one that makes your text more emoji-y, and one that gives you a random word every 10 words.
Demonstrate that an LLM can't already perform these operations by prompting it and showing the disappointing results.
Implement your two actions. Share the code in your post.
@mcp.tool()
def emojify(text: str) -> str:
"""a tool that turns text into emojis. For example, "hello world" -> "👋🌍"."""
words = text.split()
res = []
for word in words:
emojified = emoji.emojize(word, language='alias')
res.append(emojified)
return "".join(res)
Add your MCP server to GitHub Copilot. Document how you connected it.
I went to my vs-code workspace settings, added an mcp entry into the json, then added the python command to run it.
Share the prompt and response for GitHub Copilot.
