README.md•5.6 kB
# 🍓 pihole-mcp-serer
A Model Context Protocol (MCP) server for Pi-hole. This server exposes Pi-hole functionality as tools that can be used by AI assistants.
## Dependencies
### Docker
- [Docker install guide](https://docs.docker.com/engine/install/)
### `uv` (Optional, for Development)
If you want to run the application locally, use `uv`. Install it with your package manager of choice.
## Environment
Create a `.env` file in the project root with your Pi-hole credentials:
```
# Primary Pi-hole (required)
PIHOLE_URL=https://your-pihole.local/
PIHOLE_PASSWORD=your-admin-password
#PIHOLE_NAME=Primary # optional, defaults to URL if unset
# Secondary Pi-hole (optional)
#PIHOLE2_URL=https://secondary-pihole.local/
#PIHOLE2_PASSWORD=password2
#PIHOLE2_NAME=Secondary # optional
# Up to 4 Pi-holes:
#PIHOLE3_URL=...
#PIHOLE3_PASSWORD=...
#PIHOLE3_NAME=...
#PIHOLE4_URL=...
#PIHOLE4_PASSWORD=...
#PIHOLE4_NAME=...
```
## Project Structure
The project follows a modular organization for better maintainability:
```
/
├── main.py # Main application entry point
├── tools/ # Pi-hole tools organized by functionality
│ ├── __init__.py
│ ├── config.py # Configuration-related tools (DNS settings)
│ └── metrics.py # Metrics and query-related tools
├── resources/ # MCP resources
│ ├── __init__.py
│ └── common.py # Common resources (piholes://, version://)
├── docker-compose.yml # Docker Compose configuration for production
├── docker-compose.dev.yml # Docker Compose for development with volume mounts
└── Dockerfile # Docker build configuration
```
This structure separates the code into logical components while maintaining compatibility with all run modes.
## Running the Server
There are several ways to run the Pi-hole MCP server:
### Using Docker (Recommended for Production)
```bash
# Standard deployment
docker-compose up -d
```
The server will be available at `http://localhost:8383`
### Development Mode with Docker
For development, use the dev compose file which builds locally:
```bash
docker-compose -f docker-compose.dev.yml up
```
### MCP Inspector
You can run the MCP inspector using `uv` and the `mcp` CLI:
```bash
uv run mcp dev main.py
```
This will start an interactive interface at `http://localhost:6274` where you can test tools and resources.
## API
This MCP server exposes the following resources and tools:
### Resources
- `piholes://`: Returns information about all configured Pi-holes
- `version://`: Returns the MCP server version
- `list-tools://`: Returns a list of tool categories
- `list-tools://{category}`: Returns a list of tools within a specific category
### Tools
Each tool call returns results as a list of dictionaries with the following structure:
```
[
{
"pihole": "Pi-hole Name",
"data": [...] # Result data from this Pi-hole
},
...
]
```
#### Configuration
- `list_local_dns`: Lists all local DNS settings from Pi-hole(s)
- `add_local_a_record`: Adds a local A record to Pi-hole(s).
- `add_local_cname_record`: Adds a local CNAME record to Pi-hole(s).
- `remove_local_a_record`: Removes all A records for a hostname.
- `remove_local_cname_record`: Removes all CNAME records for a hostname.
#### Metrics
- `list_queries`: Fetches the recent DNS query history from Pi-hole(s)
- `list_query_suggestions`: Get query filter suggestions
- `list_query_history`: Get activity graph data for queries over time
## Testing in `goose`
Goose is a CLI LLM client that's useful for testing and development. Follow their install instructions [here](https://block.github.io/goose/docs/quickstart/).
The following assumes you've completed the initial setup with `goose configure`.
### Configure Extension
1. Type `goose configure` to open the configuration menu.
2. Select **Add Extension**
3. Select **Remote Extension**
4. It will ask for a name. It doesn't matter what you name it. I called mine `pihole-mcp`.
5. When it asks _"What is the SSE endpoint URI?"_ enter `http://localhost:8383/sse`.
6. Enter a timeout.
7. Add a description if you'd like.
8. Select **No** when it asks about environment variables.

### Start a Session
Once the server is installed, start a chat session.
```sh
goose session
```
Try asking it: _"What are my local DNS records?"_

...or telling it: _"Show me my recent DNS queries."_

## Claude Desktop
Claude's desktop client currently only support's the STDIO protocol, however you can use a proxy to communicate with the SSE endpoint.
Add the following to your `claude_desktop_config.json` file.
```json
{
"mcpServers": {
"pihole": {
"command": "npx",
"args": [
"mcp-remote",
"http://localhost:8383/sse"
]
}
}
}
```
If you're connecting to a different host on your local network and using an unsecured connection, you'll need to explicitly allow it with the `--allow-http` argument. For example:
```json
{
"mcpServers": {
"pihole": {
"command": "npx",
"args": [
"mcp-remote",
"http://192.168.1.255:8383/sse",
"--allow-http"
]
}
}
}
```
Afterwards, completely restart the application and try it out.


## License
[MIT](./LICENSE)