# TransformerBee.MCP

This is a simple PoC of a Model Context Protocol (MCP) server for [transformer.bee](https://github.com/enercity/edifact-bo4e-converter/), written in Python.
Under the hood it uses [`python-mcp`](https://github.com/modelcontextprotocol/python-sdk) and [`transformerbeeclient.py`](https://github.com/Hochfrequenz/TransformerBeeClient.py).

## Installation
You can install the MCP server as Python package or pull the Docker image.
### Install as Python Package
```shell
uv install transformerbeemcp
```
or if you are using `pip`:
```sh
pip install transformerbeemcp
```
### Install as Docker Image
```sh
docker pull ghcr.io/hochfrequenz/transformerbee.mcp:latest
```

## Start the Server via CLI 
### Python
_The package ships a simple CLI argument to start the server.
In a terminal **inside the virtual environment in which you installed the package (here `myvenv`)**, call:

```sh
(myvenv) run-transformerbee-mcp-server
```
### Docker
```sh
docker run --network host -i --rm -e TRANSFORMERBEE_HOST=http://localhost:5021 ghcr.io/hochfrequenz/transformerbee.mcp:latest
```
(For the environment variables `-e ...`, see below or the `transformerbeeclient.py` docs.)

## Register MCP Server in Claude Desktop
### If you checked out this repository
```sh
cd path/to/reporoot/src/transformerbeemcp
fastmcp install server.py
```
### If you installed the package via pip/uv
Modify your `claude_desktop_config.json` (that can be found in Claude Desktop menu via "Datei > Einstellungen > Entwickler > Konfiguration bearbeiten"):
```json
{
  "mcpServers": {
    "TransformerBee.mcp": {
      "command": "C:\\github\\MyProject\\.myvenv\\Scripts\\run-transformerbee-mcp-server.exe",
      "args": [],
      "env": {
        "TRANSFORMERBEE_HOST": "http://localhost:5021",
        "TRANSFORMERBEE_CLIENT_ID": "",
        "TRANSFORMERBEE_CLIENT_SECRET": ""
      }
    }
  }
}
```
where `C:\github\MyProject\.myvenv` is the path to your virtual environment where you installed the package and `localhost:5021` exposes transformer.bee running in a docker container.
Alternatively, if you haven't configured this handy CLI command
https://github.com/Hochfrequenz/TransformerBee.mcp/blob/c0898769670469df13f23b57a55fe4b71ed9795b/pyproject.toml#L101-L102

you can just call python with non-empty args.

Note that a lot of MCP tooling assumes you have [`uv`](https://docs.astral.sh/uv/) installed, so you might need to install it in your virtual environment as well.

For details about the environment variables and/or starting transformer.bee locally, check [`transformerbeeclient.py`](https://github.com/Hochfrequenz/TransformerBeeClient.py) docs.

### If you installed the package via Docker
```json
{
  "mcpServers": {
    "TransformerBee.mcp": {
      "command": "docker",
      "args": [
        "run",
        "--network",
        "host",
        "-i",
        "--rm",
        "-e",
        "TRANSFORMERBEE_HOST=http://localhost:5021",
        "ghcr.io/hochfrequenz/transformerbee.mcp:latest"
      ],
      "env": {
        "TRANSFORMERBEE_HOST": "http://localhost:5021",
        "TRANSFORMERBEE_CLIENT_ID": "",
        "TRANSFORMERBEE_CLIENT_SECRET": ""
      }
    }
  }
}
```
I'm aware, that using the `--network host` option is a bit hacky and not best practice.