# Deploying the Proto-OKN Unified MCP Server on AWS
## What It Is
A lightweight Python server that exposes 27 Proto-OKN knowledge graphs through a single MCP endpoint. It proxies SPARQL queries to the FRINK platform (`frink.apps.renci.org`) — it doesn't host any data, so resource requirements are minimal (a `t3.small` or equivalent is fine).
## Setup
```bash
# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh
# Clone and install
git clone https://github.com/sbl-sdsc/mcp-proto-okn.git
cd mcp-proto-okn
uv sync
```
Optional: set `MCP_PROTO_OKN_API_KEY=<token>` for Bearer-token authentication.
## Starting the Server
Three equivalent ways to start the server:
```bash
# Option 1: uv run (manages venv automatically, no install needed)
uv run mcp-proto-okn-unified --transport streamable-http --port 8000
# Option 2: Installed console script (after uv sync or pip install -e .)
mcp-proto-okn-unified --transport streamable-http --port 8000
# Option 3: Direct Python module invocation
python -m mcp_proto_okn.unified_server --transport streamable-http --port 8000
```
## HTTPS
Claude Desktop requires HTTPS with a valid domain. Put a reverse proxy (Caddy, nginx, etc.) in front of the server with a TLS certificate, and point a DNS record at the instance. Claude Code can use plain HTTP for dev/testing.
## Client Configuration
```json
{
"mcpServers": {
"proto-okn": {
"type": "url",
"url": "https://your-domain.example.com/mcp"
}
}
}
```
## For Production
Use systemd (or equivalent) to keep the process running across reboots. Alternatively, containerize it and run on a managed service like ECS Fargate or Cloud Run with a load balancer handling TLS.
