FastMCP on Wojciech Marusiak IT Blog

Giving AI Access to Your NAS: Building a QNAP MCP Server for Disk Health & Backup Audits

Wed, 21 Oct 2026 10:00:00 +0200

QNAP QTS is a powerful operating system for enterprise and homelab storage, but navigating the web interface just to check drive health, QuLog alerts, or HBS3 replication jobs can be slow and tedious.

In this post, I detail how we built and deployed mcp-qnap—a FastMCP microservice listening on port 8442 that allows Antigravity to perform deep diagnostics, SMART disk inspections, and shared folder ACL audits directly from the prompt.

The Architecture: FastMCP on QTS

We interface with QNAP’s underlying QTS management API using Python and FastMCP. The service exposes read-only endpoints secured with Bearer token authentication:

Endpoint: http://10.10.30.25:8442/mcp
Transport: HTTP JSON-RPC
Auth: Pre-shared API Key in Authorization: Bearer <token>

What the AI Can Inspect

Our mcp-qnap server registers 25 granular inspection tools:

1. Storage & SMART Telemetry

list_disks / list_storages: Checks NVMe and HDD SMART temperatures, reallocated sector counts, and pool capacity.
get_pool_usage_history / get_qts_volume_usage_history: Identifies capacity growth trends.

2. Access Control & Shared Folders

list_shared_folder / get_shared_folder_permission_acl: Audits SMB/NFS permissions and subfolder ACL inheritance.
verify_shared_folder_permission: Verifies whether user wojcieh or service account kopia has write access without guessing.

3. System Load & Security

query_top_processes / query_load_avg: Diagnoses CPU/RAM spikes without SSH logins.
get_securitycenter_report / get_qvr_logs: Scans for failed login attempts and firewall blocks.

Real-World Example: Automating 3-2-1 Backup Verification

When verifying our automated Kopia backups to QNAP NFS:

User: "Antigravity, audit my QNAP backup storage pool before we trigger the weekly offsite sync."

Agent:
1. Calls mcp_qnap_list_disks: All 4 WD Red Pro drives healthy, max temp 36°C, 0 bad sectors.
2. Calls mcp_qnap_get_shared_folder(folder="Kopia-Backups"): 2.4 TB used, 4.1 TB free.
3. Calls mcp_qnap_verify_shared_folder_permission(user="kopia-svc", folder="Kopia-Backups"): Write Access: OK.

Agent Response:
"Storage Pool 1 is healthy (all drives nominal at 36°C). The Kopia-Backups share has 4.1 TB available space with correct POSIX write permissions. Ready for offsite sync."

By connecting your NAS via MCP, your AI assistant becomes a proactive storage reliability engineer.

Controlling UniFi Networks with AI: Building a Custom UniFi MCP Server for Topology & Diagnostics

Wed, 14 Oct 2026 10:00:00 +0200

Troubleshooting complex homelab networking issues (VLAN misconfigurations, AP channel congestion, rogue IoT devices, PoE budget limits) usually requires clicking through half a dozen sub-menus in the UniFi Network Controller.

In this post, I explain how we built a containerized UniFi MCP Server that connects your UniFi Controller to AI pair-programming agents like Antigravity, giving the agent live introspection and diagnostic control over your entire network.

Why Expose UniFi to AI via MCP?

Instead of manually logging into 192.168.1.1 to check client lists, an MCP-connected agent can:

Instant Client Lookup: “Find IP, MAC, and connected switch port for device esp32-box-voice.”
VLAN & Port Audits: “List all switch ports configured with VLAN 30 (IoT) and check their link speeds.”
AP Radio Telemetry: “Which Access Point has the highest channel utilization on 5GHz right now?”
Batch Operations: Execute safe batch operations across multiple switches with structured tool confirmations.

Docker Deployment: `mcp-unifi`

We run mcp-unifi as a lightweight container on our primary Docker host (10.10.20.15):

services:
  mcp-unifi:
    image: ghcr.io/homelab/mcp-unifi:latest
    container_name: mcp-unifi
    restart: unless-stopped
    ports:
      - "32000:32000"
    environment:
      - UNIFI_HOST=https://192.168.1.1
      - UNIFI_USER=mcp-agent
      - UNIFI_PASS=${UNIFI_PASSWORD}
      - UNIFI_SITE=default
      - UNIFI_VERIFY_SSL=false
      - MCP_PORT=32000
    networks:
      - homelab-net

Exposed Tool Definitions

The UniFi MCP server exposes 5 high-impact tools to the agent:

unifi_tool_index: Returns a comprehensive catalog of all network query methods.
unifi_execute: Runs targeted API endpoints against the controller (e.g. stat/sta, stat/device, rest/networkconf).
unifi_batch: Executes sequential multi-device commands (e.g., restarting PoE on specific ports).
unifi_load_tools: Dynamically generates schema definitions for deep network querying.

Example Agent Diagnostic Session:

User: "Antigravity, why is my Home Assistant camera stream lagging?"

Agent:
-> Calling mcp-unifi: unifi_execute(endpoint="stat/sta", filter={"mac": "dc:a6:32:..."})
<- Returns: RSSI -84 dBm, TX Rate 12 Mbps, Channel 36 (Congestion 78%)

Agent Response:
"The camera is experiencing poor WiFi signal (-84 dBm) and high channel congestion (78%) on AP-LivingRoom. Moving it to AP-Hallway on Channel 149 will restore full 60fps throughput."

Having your network topology available to your AI assistant turns routine network troubleshooting into a single conversational query.