My AI Product ③
The previously mentioned JAA has been operating continuously for six months and has gained a stable user base. Tomorrow, I need to attend the company's OPEN AI DAY to share about JAA in Shanghai.
If it were half a year ago, I would have been very willing and had a lot to say, but now, I think it's already somewhat outdated.
After JAA had been running for a short while, I once thought about what other directions it could develop in:
First, the core of JAA is function call, so simply replacing the tools can turn it into a Jenkins AI Assistant, GitHub AI Assistant, and so on.
Second, perhaps I can package the tool part of my code into a Python library. Then others can download this library and develop applications that go beyond my existing framework (I'm using AutoGen).
While I was having these ideas (early December 2024), Anthropic introduced MCP at the end of the previous month.
So I immediately went to GitHub to create a repo named mcp-jira,
but I discovered that mcp-atlassian (Jira + Confluence) already existed and had 100 stars.
Instead, I contributed some code to that repository and discussed a few ideas with the author via GitHub.
At the same time, since I also needed to use Jenkins for work, I developed another project called mcp-jenkins. Although there were other similar repo at the time, they had few stars, limited tool support, and I felt their engineering practices were lacking.
After developing my own MCP server, I abandoned the initial idea of replacing function calls with MCP.
I realized that MCP is better suited for providing general-purpose functionalities, while my JAA contains a lot of independent business logic that might change with API versions. From both a security and practicality perspective, it was not suitable to serve as an MCP server.