A framework for building MCP servers.
And a growing catalog of production-ready tool packages.
Build MCP servers in minutes. Every tool package follows the same pattern.
The foundation. Config loading, transport (stdio + HTTP), health endpoint, OAuth, plugin system. Everything you need to build an MCP server.
$ pip install mcp-server-framework
# Your plugin:
def register(mcp, config):
@mcp.tool()
def my_tool(x: str):
return f"done: {x}"
Every tool package works as a standalone MCP server and as a regular Python library. No MCP required to use the functions.
# As MCP server
$ mcp-shell-tools
# As Python library
from mcp_shell_tools.shell import tools
result = tools.file_read("/etc/hosts")
Implement register(mcp, config). Add @mcp.tool() decorated functions. Your package works standalone and as a framework plugin.
# Claude Desktop config
{
"mcpServers": {
"shell": {
"command": "mcp-shell-tools"
}
}
}
One interface. Standalone or plugin. MCP server or Python library.
Each tool package is on PyPI. One pip install, ready to use.
$ pip install mcp-shell-tools
$ mcp-shell-tools
✓ 26 tools, stdio ready
Every tool works as a Python function. No MCP needed.
from mcp_shell_tools.shell import tools
result = tools.file_read("/etc/hosts")
files = tools.glob_search("**/*.py")
The framework handles transport, config, health. You write the tools.
def register(mcp, config):
@mcp.tool()
def greet(name: str):
return f"Hello, {name}!"
One function. That's the entire plugin API.
Yes. Every package exports pure Python functions. Import and call directly — no MCP, no server, no overhead.
Implement register(mcp, config). That's the entire API. Add @mcp.tool() decorated functions inside. Your package works standalone and as a framework plugin.
Yes. Every tool package runs as a stdio MCP server. Add it to your Claude Desktop config and it just works.
Python 3.10 or newer. All packages are tested on 3.10, 3.11, and 3.12.
Each package works standalone or as a framework plugin. Install, run, done.
26 tools — Filesystem, editor, search, shell, system diagnostics. The workstation toolkit for AI agents.
pip install mcp-shell-tools
60 tools — HomeMatic CCU3 / OpenCCU smart home. Full JSON-RPC API coverage: devices, channels, rooms, programs, system variables.
pip install mcp-homematic-tools
43 tools — Browser automation. Navigation, interaction, content extraction, screenshots, semantic locators.
pip install mcp-playwright-tools
6 tools — Image processing. Read, resize, crop, convert, screenshot. Base64 output for direct use in LLM context.
pip install mcp-image-tools