[Reference Link:]
1,
https://mp.weixin.qq.com/s?__biz=MzAwNDEyNTg0MA==&mid=2649952903&idx=1&sn=c2345968688b792713836a19ba9a0a68&chksm=82c9cc1a91ce6fc113f909f77bed555a42eea8913fa2cb24e4cf1045b7a434a7155120a3bcda&mpshare=1&scene=2&srcid=0106KLCAc4ucqaMZHSFSpMmw&sharer_shareinfo=02088532e517bf4cb86ea447109b83b4&sharer_shareinfo_first=7f2b5d39c0d49ce7c5d6d19394eaf41b#rd
It’s simpler than a button! Quickly build an Agent workflow with Cline + MCP: Practical case analysis
Generated foreign media comments for military bloggers directly using Cline to create the MCP server
2,
https://openrouter.ai/
OpenRouter itself seems to be a website aggregating LLMs and providing a unified interface, one of its main values is actually a ranking based on token consumption
3,
https://github.com/RooVetGit/Roo-Cline
Roo-Cline, a fork version of the Cline plugin, with more new features
4,
https://mp.weixin.qq.com/s?__biz=MzI5MjQ3ODY3Mw==&mid=2247491310&idx=1&sn=9a327156e3be7e5d2b583ede24a6a3cc&chksm=edacb28426140879c9ac778528e9c9d5eb58b6a95d23c4741ec9af45027a287606f3b6e06e5f&mpshare=1&scene=2&srcid=0106kz7MnKoKWUSsGcyHjXos&sharer_shareinfo=8a159cbaf6975e190257207ee8521ce2&sharer_shareinfo_first=8a159cbaf6975e190257207ee8521ce2#rd
[Test] Is DeepSeek V3 really that amazing? Partnering with Roo Cline, comparing programming capabilities of Claude and o1
=========================
[Installation]
It’s quite simple, both are available in vscode
First, install the official version
Then the sidebar will have a little robot icon
Then:
Here in API Provider, DeepSeek is directly available, officially supported
Many previous articles seem outdated
Then it requires an API key, just jump to the official website
100 yuan for 50 million tokens
Then go to
Remember to save the key
Then go back to vscode
Drag the little robot to the right, so you can get a left-center-right layout
It seems that a long monitor is needed in the AI era
=============================================
[Practice]
First, create a new folder
You must cut out the auxiliary bar to see Cline now
1. Try creating the first MCP Agent
Create a MCP server that can download the transcript of a YouTube video when the video’s URL is given.
The result is that a json configuration was created first to configure the MCP server
Then a ts file was created
Then the tool itself was also configured
Next, let’s experiment with the tool itself:
get transcript of https://www.youtube.com/watch?v=GBR6pHZ68Ho
Error says the service is not started
It requires manual startup
The service has started, I can work now
Is it OK?
It asked me to restart vscode
Alright, restart
I really couldn’t stand it, uninstalled Node 18
Installed the latest 22
Then the latest version of Node.js actually upgraded Python to 3.13 first
https://mp.weixin.qq.com/s/d5kCwHOdpXZTzlJuNXKVwg?poc_token=HG7BfGejyk7GFP5b4Mq4wL9icbRoHIGd-Ydbf7AA
A little tip
Switch the terminal to bash
========================================
[Improvements]
I was stuck all afternoon on various errors in the generated MCP SERVER code, and finally found that the problem was still due to the code written by DeepSeek not knowing what the MCP standard was at that time
I changed my approach
https://github.com/modelcontextprotocol/servers/tree/main/src/fetch
Installed an official similar example library
Then placed it in the same level directory
Let the LLM analyze it
Then gave it a key instruction:
Let it reference this project to rewrite weather_server.py
After it finished rewriting, it worked instantly, the green light lit up
Ah, it took me an entire afternoon!! Finally succeeded…
So it’s still necessary to let AI do the homework, relying on its own knowledge base is useless