Hello World to Model Context Protocol
AI agents are hot topics these days, and one of the most important capability for AI agents is tooling (imagine that for a brain to work, it needs to have access to external information (e.g. receive, which can be achieved through touching, smelling, seeing, hearing, etc., action: touching, blowing, etc.). In AI agents world, you need both read/write access (some people like to call them knowledge and action). Industry is moving towards providing AI agent with more tool use capabilities: 1, 2.
With so many data/information around the world, we need to provide agents the standardized access to external data sources and tools (e.g. langchain/crewai has own way to define tool, same with llamaindex, pydantic ai), so AI systems know how to talk to external data sources and tools in a unified way. MCP (model context protocol) is an open-source project released by Anthropic in this direction.
Currently to test this, we need to use local desktop. As in MCP user quickstart, it says
Why Claude for Desktop and not Claude.ai?
Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
To try this, follow the guide 1, 2. Prerequisite:
config
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-filesystem",
"C:\\Users\\<username>\\Documents",
"C:\\Users\\<username>\\Downloads"
]
}
}
}After modifying config, you need to close all Claude processes (e.g. end task in Task manager) to let it pick up new config.
Appendix
- Frontend Development: Integrate AI with component libraries to streamline UI/UX design, allowing more intuitive and responsive user experiences.
- 3D Design: Use AI to create 3D models or scenes in tools like Blender, guided by natural language prompts for enhanced creativity and efficiency.
- Browser Automation: Automate repetitive browser tasks, such as web scraping or testing, using tools like Puppeteer to save time and reduce manual effort.
- Database Management: Enable natural language queries and efficient application development with tools like Subase, simplifying complex database interactions.
Used file system, brave search, github MCP servers
Brief walkthrough of MCP server, client SDK using Puppeteer browser automation
cursor IDE and MCP
openai agent sdk
