The terminology of MCP (Model Context Protocol, for AI agents & tools communication) is slightly confusing.
The "Host" is the app that has AI/LLM functionality, and may need to communicate with external
resources, tools or even prompts.
The "MCP client" a custom app or a stand-alone app that communicates with MCP "server"
The "MCP server" is usually a lightweight "proxy" app that provide requested resource or invokes tools/functions on demand of MCP client.
The communication between of MCP client & servers can be
- stdio: when "server" app is started as a process, with input/output/error redirected to "client" app
that used to be a common on desktop apps, and is relatively simple and efficient for local "server apps" - SSE: HTTPS based "streaming"; relatively standard web server API app
Core architecture - Model Context Protocol
What are MCP host apps (with client plugins)
- Cursor.ai IDE app
- Claude.ai desktop app Download - Claude
- Cline - AI Autonomous Coding Agent for VS Code
MCP client apps
There is already many MCP servers created and made available.
And there are "online marketplaces" for finding those servers
MCP Market | Discover Top MCP Servers //mcpmarket.com
MCP Servers //mcp.so
Find Awesome MCP Servers and Clients | MCPFly //mcpserver.so
Servers | Model Context Protocol | MCP //model-context-protocol.com/servers
some example MCP servers
Caution
MCP servers can run arbitrary code on your machine. Only add servers from trusted sources, and review the publisher and server configuration before starting it.
- AWS KB Retrieval - Retrieval from AWS Knowledge Base using Bedrock Agent Runtime
- Brave Search - Web and local search using Brave's Search API
- EverArt - AI image generation using various models
- Everything - Reference / test server with prompts, resources, and tools
- Fetch - Web content fetching and conversion for efficient LLM usage
- Filesystem - Secure file operations with configurable access controls
- Git - Tools to read, search, and manipulate Git repositories
- GitHub - Repository management, file operations, and GitHub API integration
- GitLab - GitLab API, enabling project management
- Google Drive - File access and search capabilities for Google Drive
- Google Maps - Location services, directions, and place details
- Memory - Knowledge graph-based persistent memory system
- PostgreSQL - Read-only database access with schema inspection
- Puppeteer - Browser automation and web scraping
- Redis - Interact with Redis key-value stores
- Sentry - Retrieving and analyzing issues from Sentry.io
- Sequential Thinking - Dynamic and reflective problem-solving through thought sequences
- Slack - Channel management and messaging capabilities
- Sqlite - Database interaction and business intelligence capabilities
- Time - Time and timezone conversion capabilities
example: chess engine MCP server (using UCI stdio)
turlockmike/chess-mcp: A MCP Server with various chess functionality @GitHub
Interact with the Lichess chess platform using natural language.
VS Code MCP client extensions
- Use MCP servers in VS Code (Preview)
- Copilot MCP - Visual Studio Marketplace
- VSCode MCP Server - Visual Studio Marketplace
MCP servers can run as public/auth webs servers
here is example setup of Claude Desktop app on Windows machine
it gets "technical" quite quickly :)
this requires enabling "developer" tools from settings (from menu options)
requires editing config file; replace <user> and <folder-path>
C:\Users\<user>\AppData\Roaming\Claude
This setup is with Python demo weather MCP server app
using Python directly without virtual envs
make sure to
> pip install FastMCP
to find file path to Python exe can use this command;
If python is in "path" env var it may work directly without full path, with just "python" command.
If python is in "path" env var it may work directly without full path, with just "python" command.
`python -c "import sys; print(sys.executable)"`
{
"mcpServers": {
"weather": {
"command": "C:\\Users\\<user>\\AppData\\Local\\Programs\\Python\\Python312\\python.exe",
"args": [
"<folder-path>\weather-server-python\\weather.py"
]
}
}
}
here we are using TypeScript version of weather demo MCP server app.
for this need to run
> npm install
> npm run build
for setting Cursor.ai desktop app
from menu options select > File > Preferences > Cursor Settings > MCP
[ + Add new global MCP server ]
the path in args should match your local file path to result of build of TypeScript app to index.js
{
"mcpServers": {
"weather": {
"command": "node",
"args": ["D:\\GitHubTest\\ai-mcp\\quickstart-mcp-weather\\weather-server-typescript\\build\\index.js"]
}
}
}
Cursor requires interactive "approval" to run MCP tool, similar to Claude desktop app.
No comments:
Post a Comment