Lite_MCP_sdk
a light weight version of mcp sdk inspired by official model context protocol sdk, NOW focus on sse only no stdio
claude mcp add --transport stdio s2thend-lite_mcp_sdk node path/to/server.js \ --env PORT="3000" \ --env EXAMPLE_CONFIG="placeholder"
How to use
Lite MCP SDK for JavaScript provides a lightweight, SSE-based MCP server and a client kit to interact with it. The server exposes a simple “tools/call” endpoint to invoke named tools with argument payloads, and the client can list available tools and execute them remotely. A minimal example shows how to define tools (e.g., a calculate-bmi tool) and how clients can connect, request tool lists, and call tools. With the SDK, you can build your own remote MCP server that communicates over SSE, enabling low-overhead, event-driven interactions between clients and server-side tools. The client side demonstrates constructing a server client, connecting to the SSE endpoint, listing tools, and calling a specific tool with arguments, returning structured responses. This makes it suitable for scenarios where you want lightweight remote computation or agent-style tool invocation over SSE with minimal dependencies.
How to install
Prerequisites:
- Node.js (16+ recommended) and npm
- Optional: a runtime environment for your chosen server entry point (e.g., a custom server.js file)
Install the Lite MCP SDK package (and any peer dependencies you plan to use, like Express or CORS):
npm install lite-mcp-sdk
npm install express
npm install cors
Typical workflow (as shown in the repository):
- Create a Node.js server that uses the Lite MCP SDK Server and defines tools. For example, require the SDK, instantiate Server with a transport (e.g., SSEServerTransport), define tools, and start listening on a port.
- On the client side, require or import Client and SSEClientTransport, configure connection to http://localhost:3000 (or your chosen port), connect, list tools, and call a tool by name with arguments.
- Run your server file with Node:
node path/to/your-server.js
The repository includes concrete code samples for both server and client usage, demonstrating a BMI tool, the remote call flow, and the transport setup.
Additional notes
Notes and tips:
- This repository emphasizes a SSE-based (Server-Sent Events) transport with no fixed zod types, focusing on lightweight validators and a simple, flexible protocol.
- If you split server and client into separate processes, ensure CORS settings (if using Express) allow your client to connect from a browser or another origin.
- The example tools are defined with a name, description, input schema, and an async handler. You can replace or extend these with your own tool implementations.
- Common pitfalls: ensure the server port is open and not blocked by firewalls; confirm the SSE endpoints (/messages for transport setup and /sse for client connection as shown in examples) are accessible; verify that your client transports are initialized with the correct URLs and headers.
- Environment variables can be used to configure PORT, host, or any runtime-specific options; document and reference them in your deployment config.
Related MCP Servers
PPTAgent
An Agentic Framework for Reflective PowerPoint Generation
paperdebugger
A Plugin-Based Multi-Agent System for In-Editor Academic Writing, Review, and Editing
multimodal-agents-course
An MCP Multimodal AI Agent with eyes and ears!
python -client
支持查询主流agent框架技术文档的MCP server(支持stdio和sse两种传输协议), 支持 langchain、llama-index、autogen、agno、openai-agents-sdk、mcp-doc、camel-ai 和 crew-ai
zerodha
Zerodha MCP Server & Client - AI Agent (w/Agno & w/Google ADK)
openai -agent-dotnet
Sample to create an AI Agent using OpenAI models with any MCP server running on Azure Container Apps