MCP Server HTTP Streamable

You can run a minimal MCP server that serves a Streamable HTTP endpoint, exposing a greeting tool you can call from MCP clients. It supports running locally or in containerized environments, and can be deployed on Kubernetes for scalable access. The server listens on HTTP and provides an easy way to connect tools into your MCP workflows over the network.

How to use

Connect MCP clients to the HTTP endpoint at http://localhost:8000/mcp by default. You can run the server locally, in Docker, or in Kubernetes, and you can use standard MCP clients to invoke the available tools.

How to install

Prerequisites to run the MCP server are: Python 3.12 or newer and the uv transport runtime. You can also use uvx to run the server from a Git source, or run the server in Docker or in Kubernetes.

# Install Python (3.12+)
# Install uv (recommended) or uv via pip
pip install uv

# Optionally install uvx to run from Git
pip install uvx

# If you plan to use Docker and Kubernetes:
# Ensure Docker and Kubernetes tooling are installed (docker, docker-compose, kubectl)

Configuration and environment

The server supports optional environment variables to customize binding. Use MCP_HOST to set the bind address (default 0.0.0.0) and MCP_PORT to set the listening port (default 8000). When you start the server with a custom port, MCP endpoints adjust accordingly.

# Example: run on a different port
MCP_PORT=9000 uv run mcp-server

Available tools

greeting

Returns a greeting for the given name.