genieacs-mcp
<p align="center">
<img src="https://raw.githubusercontent.com/GeiserX/genieacs-mcp/main/docs/images/banner.svg" alt="genieacs-mcp" width="100%">
</p>
<p align="center">
<a href="https://www.npmjs.com/package/genieacs-mcp"><img src="https://img.shields.io/npm/v/genieacs-mcp?style=flat-square&logo=npm" alt="npm"></a>
<a href="https://github.com/GeiserX/genieacs-mcp/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/GeiserX/genieacs-mcp/ci.yml?style=flat-square&logo=github&label=CI" alt="CI"></a>
<a href="https://github.com/GeiserX/genieacs-mcp/blob/main/LICENSE"><img src="https://img.shields.io/github/license/GeiserX/genieacs-mcp?style=flat-square" alt="License"></a>
<a href="https://hub.docker.com/r/drumsergio/genieacs-mcp"><img src="https://img.shields.io/docker/pulls/drumsergio/genieacs-mcp?style=flat-square&logo=docker" alt="Docker Pulls"></a>
<a href="https://github.com/GeiserX/genieacs-mcp/stargazers"><img src="https://img.shields.io/github/stars/GeiserX/genieacs-mcp?style=flat-square&logo=github" alt="GitHub Stars"></a>
</p>
**genieacs-mcp** is an MCP server for [GenieACS](https://github.com/genieacs/genieacs), the open-source TR-069 ACS that manages routers, ONTs and other CPE devices. Add it to Claude Desktop, Cursor or any other MCP client and ask in plain words: which devices stopped informing last night, what firmware serial 000003 runs, reboot the ones tagged `pilot`. The assistant reads and acts through your GenieACS NBI with 7 resources and 12 tools; one Go binary, no database of its own.
<p align="center">
<img src="https://raw.githubusercontent.com/GeiserX/genieacs-mcp/main/docs/images/screenshots/search-devices.png" alt="MCP Inspector connected to genieacs-mcp: the search_devices tool run with the query _tags residential, and the result listing simulated Huawei BM632w devices with their ids, manufacturer, last inform and tags" width="100%">
</p>
## Features
- Find devices the way you would ask a colleague: by tag, model, firmware, last inform, or any TR-069 parameter (`search_devices` takes a MongoDB-style filter).
- Read a device's full parameter tree, its tasks and its faults, plus the presets, provisions and files on the ACS, as resources (`genieacs://device/{id}`, `genieacs://devices/list`, ...).
- Reboot a device, push firmware to it, refresh or set a parameter, or wake it with a connection request, from a chat.
- Tag devices, create or delete presets and provisions, and delete or retry tasks, so the assistant can fix what it finds.
- Every tool description says what it does, its limits and an example, so clients pick the right one without a prompt from you.
- Read versus act: `search_devices`, `get_parameter` and the seven resources only read; the other ten tools queue a task on a device or change the ACS.
- No read-only mode: point it at an ACS you are willing to let an assistant act on, and keep your client's tool-approval prompts on.
- Safe on a laptop: the HTTP transport listens on loopback only unless you set an address and a bearer token, and checks `Host` and `Origin` against DNS rebinding.
- One static Go binary for Linux, macOS and Windows on amd64 and arm64, also as an npm package (`npx genieacs-mcp`) and a Docker image.
## Quick start
You need a GenieACS whose NBI (port 7557) this machine can reach; without one, the family's demo stack gives you an ACS with a simulated router in about a minute:
```sh
curl -fsSLO https://raw.githubusercontent.com/GeiserX/genieacs-container/main/docker-compose.yml
docker compose --profile testing up -d # GenieACS + MongoDB + one simulated CPE, NBI on localhost:7557
```
Then add the server to a client that reads an `mcpServers` block (Claude Desktop, Cursor, Claude Code; in VS Code the `.vscode/mcp.json` key is `servers`):
```json
{
"mcpServers": {
"genieacs": {
"command": "npx",
"args": ["-y", "genieacs-mcp"],
"env": { "ACS_URL": "http://localhost:7557" }
}
}
}
```
It worked when you ask "which devices does the ACS know about?" and the assistant answers with device ids such as `202BC1-BM632w-000000` (the demo stack) or your own. Docker, the HTTP transport and a local build are in [Getting started](https://geiserx.github.io/genieacs-mcp/getting-started/).
## Documentation
The full documentation is at [geiserx.github.io/genieacs-mcp](https://geiserx.github.io/genieacs-mcp/).
- [Getting started](https://geiserx.github.io/genieacs-mcp/getting-started/): npm, Docker Compose, local build, and what the first exchange looks like
- [Configuration](https://geiserx.github.io/genieacs-mcp/configuration/): every environment variable, the HTTP transport's security model, client config for stdio and HTTP
- [Usage](https://geiserx.github.io/genieacs-mcp/usage/): the 7 resources and 12 tools with their arguments, which ones act on devices, example prompts
- [Development](https://geiserx.github.io/genieacs-mcp/development/): tests, trying the server with MCP Inspector, contributing
- [Related projects](https://geiserx.github.io/genieacs-mcp/related/): the GenieACS family, other MCP servers, where this server is listed
## Related projects
Listed in the [official MCP Registry](https://registry.modelcontextprotocol.io) as `io.github.GeiserX/genieacs-mcp`. Part of the GenieACS family: [genieacs-container](https://github.com/GeiserX/genieacs-container) (the ACS itself, Docker and Helm, with this server as a compose profile), [genieacs-sim-container](https://github.com/GeiserX/genieacs-sim-container), [genieacs-ansible](https://github.com/GeiserX/genieacs-ansible), [genieacs-ha](https://github.com/GeiserX/genieacs-ha), [genieacs-services](https://github.com/GeiserX/genieacs-services), and [n8n-nodes-genieacs](https://github.com/GeiserX/n8n-nodes-genieacs) (archived).
## License
[GPL-3.0-or-later](https://github.com/GeiserX/genieacs-mcp/blob/main/LICENSE)
TDQS
Scored across 12 tools
Each tool targets a distinct operation: device wake-up, reboot, firmware download, parameter set/refresh/read, device search/tag, preset/provision management, and task delete/retry. No overlapping purposes, clear differentiation.
All tools follow a consistent verb_noun pattern (e.g., delete_task, set_parameter, tag_device). No mixing of styles or ambiguous verbs.
12 tools cover the core operations for TR-069 device management: device interaction, parameter handling, task management, device discovery, and configuration management. Well-scoped without being excessive.
Covers essential CRUD and lifecycle operations for devices, parameters, tasks, presets, and provisions. Minor gap: no dedicated tool to list all queued tasks, but task management tools (delete, retry) and reference to task resources partially compensate.