mcp-tool-kit
Agentic abstraction layer for building high precision vertical AI agents written in python for Model Context Protocol.
claude mcp add --transport stdio getfounded-mcp-tool-kit docker run -i --name mcp-toolkit ghcr.io/getfounded/mcp-tool-kit:latest
How to use
MCP Tool Kit is a comprehensive toolkit for the Model Context Protocol that automatically discovers tools at runtime and exposes them through multiple transports. It supports stdio for desktop integrations and Server-Sent Events (SSE) for web access, giving you a flexible way to interact with more than 120 built-in and pluggable tools (such as filesystem access, time tools, web search, browser automation, data analysis, and document manipulation). The toolkit is designed to be configured with a simple config.yaml and environment variables for API keys and other secrets. Once running, tools are registered dynamically, so you can add or customize capabilities without heavy configuration. You can connect to the server via SSE endpoints or through a standard docker/pipeline setup, depending on your deployment.
To use Claude Desktop or other clients, run the server in the desired mode (stdio or SSE) and point your client to the proper transport. The toolkit exposes a unified interface for commands across tools, and you can customize per-tool behavior with the provided configuration blocks. The documentation site covers details on tool availability, configuration options, and examples for integrating with external services like Brave Search, FRED, Yahoo Finance, and more. The example shows how to reference the server in Claude Desktop by embedding a docker-based execution path for the MCP server.
How to install
Prerequisites:
- Docker installed and running on your host (or permissions to run Docker commands)
- Optional: curl, wget, and shell environments for the installation method you choose
Installation methods (choose one):
- Docker (recommended):
- Prereqs: Docker installed
- Run: docker run -it --name mcp-toolkit ghcr.io/getfounded/mcp-tool-kit:latest
- Notes: This uses the official image and sets up the MCP Tool Kit server with defaults. You can connect to it via the provided transports (stdio or SSE) depending on your configuration.
- Quick setup via official install script (no Git required):
- Windows PowerShell: irm https://raw.githubusercontent.com/getfounded/mcp-tool-kit/main/install.ps1 | iex
- macOS/Linux: bash <(curl -s https://raw.githubusercontent.com/getfounded/mcp-tool-kit/main/install.sh)
- This installer will download MCP Tool Kit, check for Docker, create shortcuts, and configure the environment automatically.
- Manual Git-based setup (for developers):
- git clone https://github.com/getfounded/mcp-tool-kit.git
- cd mcp-tool-kit
- Follow the repository's developer guidance to run the launcher or start the server in your environment.
Additional setup:
- Copy .env.template to .env and populate API keys and secrets as needed (e.g., BRAVE_SEARCH_API_KEY, FRED_API_KEY, etc.).
- If using Docker, ensure you map any required volumes for storage as described in the documentation.
Additional notes
Tips and common considerations:
- Tools are discovered at runtime; no manual per-tool configuration is required for basic functionality, but you can customize via config.yaml to enable/disable specific tools and set per-tool options.
- When using SSE, you can access the server at http://localhost:8080 and use endpoints like /sse and /messages for communication.
- If you plan to connect with Claude Desktop, use the provided example in the README to reference the MCP server inside a Docker container. Adjust the command path if your deployment differs.
- Storage for file operations is isolated to a dedicated directory (e.g., ~/mcp-tool-kit-storage); ensure appropriate permissions and backup strategies for your environment.
- If you encounter issues with API keys or external services, verify that environment variables are loaded correctly (often via a .env file) and that network access to the APIs is allowed in your deployment environment.
Related MCP Servers
PPTAgent
An Agentic Framework for Reflective PowerPoint Generation
web-agent-protocol
🌐Web Agent Protocol (WAP) - Record and replay user interactions in the browser with MCP support
ContextPods
Model Context Protocol management suite/factory. An MCP that can generate and manage other local MCPs in multiple languages. Uses the official SDKs for code gen.
knowledgebase
BioContextAI Knowledgebase MCP server for biomedical agentic AI
qmt
基于 QMT 平台股票行情的 MCP 服务器,用于提供股票市场数据下载和查询的功能。
mcp-poisoning-poc
This repository demonstrates a variety of **MCP Poisoning Attacks** affecting real-world AI agent workflows.