convez.ai
MCP ServerDevOps & InfraestruturaIntermediário
HeadNerd-Jay

N-sight AI Connect — MCP Server

HeadNerd-Jay/N-sight-MCP-Server

Acesso via IA aos dados live do N-sight RMM

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que fornece acesso direto e seguro aos dados do seu ambiente N-sight RMM através de um assistente de IA. Permite consultar informações sobre clientes, dispositivos, checks e hardware em linguagem natural, sem precisar copiar dados manualmente de dashboards.

Instalação

$npx convez-ai HeadNerd-Jay/N-sight-MCP-Server

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
1
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Consultar status de checks falhando em tempo real
  • Recuperar detalhes de hardware de dispositivos específicos
  • Listar software instalado em máquinas
  • Gerar resumos de ambiente para clientes MSP
  • Automação de consultas RMM com linguagem natural

Veredito

MCP server bem estruturado para profissionais de TI e MSPs que usam N-sight RMM. Oferece integração direta com 36 ferramentas e suporta múltiplos clientes AI (Claude, ChatGPT, Copilot Studio), embora exija familiaridade com conceitos de MCP e credenciais API N-sight.

README

A tradução para o português deste README ainda está na fila.

N-sight AI Connect — MCP Server

Audience: IT Professionals and MSPs Assumes: Familiarity with MCP concepts. You know what a Model Context Protocol server is and roughly how it connects to an AI client.


What This Is

N-sight AI Connect is an MCP server that gives your AI assistant direct, read-only (or read/write) access to your N-sight RMM environment. Instead of copying data out of dashboards and pasting it into a chat window, you ask questions in plain language and the AI queries your live N-sight data on your behalf.

Examples of what you can ask:

  • "Which of my clients have failing checks right now?"
  • "Show me the hardware details for CSP-0009 at Company 1."
  • "What software is installed on AM4-MacOSx-Sequoia at Kelltic Cider?"
  • "Give me an environment summary for Kelltic Cider Company."

Documentation

GuideWhat it covers
This fileInstallation, client setup, tool reference, testing, troubleshooting
CLIENT-SETUP-GUIDE.mdDetailed setup for Claude Desktop, ChatGPT Desktop, and OpenAI Agents SDK
COPILOT-STUDIO-GUIDE.mdConnecting to Microsoft Copilot Studio via HTTP transport (v2)
TOOL-REFERENCE.mdComplete tool reference with all 36 tools, keywords, and triggers

Before You Start

You need three things:

  1. An N-sight account with API access enabled.
  2. Your N-sight API key — generate this from your N-sight portal under Settings > API.
  3. Your N-sight server URL — confirm yours in your N-sight portal.

Server URL by region:

RegionURL
North Americahttps://www.systemmonitor.us
Americas (alternate)https://www.am.remote.management
Europehttps://www.systemmonitor.eu
Asia Pacifichttps://wwwasia.systemmonitor.us
UKhttps://www.systemmonitor.co.uk

You also need Node.js 18 or later installed. Check with:

node --version

If you don't have it:

# Windows (via Chocolatey)
choco install nodejs

# Or download directly from nodejs.org

Installation

Step 1 — Get the server files

Clone the repo or download the release package:

git clone https://github.com/HeadNerd-Jay/N-sight-MCP-Server.git
cd N-sight-MCP-Server

Step 2 — Install dependencies and build

npm install
npm run build

This compiles the TypeScript source into dist/. You should see no errors.

Step 3 — Create your credentials file

In the project root, create a file named .env:

NSIGHT_API_KEY=your_api_key_here
NSIGHT_SERVER_URL=https://www.systemmonitor.us

Replace both values with your actual credentials. This file is gitignored and will never be committed to version control.

Do not put your API key anywhere else. Not in config files, not in environment variables passed through your AI client config, not in the command line. The .env file is the only place it belongs.


Connecting to Your AI Client

Claude Desktop (Windows)

Config file location:

C:\Users\<your-username>\AppData\Roaming\Claude\claude_desktop_config.json

Open that file (create it if it doesn't exist) and add the following inside the mcpServers object:

{
  "mcpServers": {
    "nsight": {
      "command": "node",
      "args": ["C:\\path\\to\\N-sight-MCP-Server\\dist\\readonly-server.js"],
      "env": {}
    }
  }
}

Replace the path with the actual full path to your dist\readonly-server.js. Use double backslashes on Windows.

To also enable write tools (clearing checks, adding notes, running tasks):

{
  "mcpServers": {
    "nsight": {
      "command": "node",
      "args": ["C:\\path\\to\\N-sight-MCP-Server\\dist\\readonly-server.js"],
      "env": {}
    },
    "nsight-production": {
      "command": "node",
      "args": ["C:\\path\\to\\N-sight-MCP-Server\\dist\\production-server.js"],
      "env": {}
    }
  }
}

Restart Claude Desktop after saving. The N-sight tools will appear automatically in the next session.

Claude Desktop (macOS)

Config file location:

~/Library/Application Support/Claude/claude_desktop_config.json

Same JSON structure as above. Use forward slashes in the path:

{
  "mcpServers": {
    "nsight": {
      "command": "node",
      "args": ["/Users/your-username/N-sight-MCP-Server/dist/readonly-server.js"],
      "env": {}
    }
  }
}

ChatGPT Desktop (Windows and macOS)

ChatGPT Desktop uses the same mcpServers JSON format. Config file locations:

  • Windows: %APPDATA%\ChatGPT\claude_desktop_config.json
  • macOS: ~/Library/Application Support/ChatGPT/claude_desktop_config.json

Use the same JSON block shown for Claude Desktop above.

You also need to enable MCP in ChatGPT Desktop settings: go to Settings > Advanced and turn on Enable MCP.

Note: MCP support in ChatGPT Desktop was introduced after Claude Desktop. If the option isn't visible, check that you're on the latest version.

OpenAI Agents SDK (Python)

For developers building agents programmatically:

from agents import Agent, Runner
from agents.mcp import MCPServerStdio

nsight_server = MCPServerStdio(
    params={
        "command": "node",
        "args": ["/path/to/N-sight-MCP-Server/dist/readonly-server.js"],
        "env": {}
    }
)

async def main():
    async with nsight_server as server:
        agent = Agent(
            name="N-sight Assistant",
            instructions="You help IT teams query their N-sight RMM environment.",
            mcp_servers=[server]
        )
        result = await Runner.run(agent, "List all my clients.")
        print(result.final_output)

Credentials are loaded from the .env file automatically by the server on startup.

Future Clients

Any MCP-compatible client that supports stdio transport can connect to this server using the same pattern:

  • Command: node
  • Args: full path to dist/readonly-server.js (or production-server.js for write access)
  • Env: empty — credentials come from .env in the project directory

If a future client requires HTTP transport instead of stdio, see COPILOT-STUDIO-GUIDE.md for setup instructions.


Available Tools — Read-Only Server

The read-only server exposes 23 tools, organized by function. Use these for monitoring, reporting, and querying without risk of changes to your environment.

Environment and Summary

ToolWhat it does
get_environment_summaryHealth and inventory snapshot for a single customer or specific sites
list_all_sitesAll sites across all clients in one call

Scope requirement: get_environment_summary requires you to specify a customer name or one or more site names. It will not run without that context and will reject any request that spans multiple customers.

Clients and Sites

ToolWhat it does
list_clientsAll clients in your N-sight account
list_sitesAll sites for a specific client

Devices

ToolWhat it does
list_devicesDevices at a specific site (requires customer or site name)
get_device_assetsAsset information for a single device

Scope requirement: list_devices requires at least a customer name or site name before it will run.

Checks and Monitoring

ToolWhat it does
list_checksMonitoring checks for a device
list_failing_checksAll checks currently failing across the account
get_check_outputDetailed output from a specific check

Patch Management

ToolWhat it does
list_patchesPatch status for a device

Software and Hardware (Single Device Only)

ToolWhat it does
list_softwareInstalled software on a device
list_hardwareHardware details for a device
list_device_asset_detailsFull asset record for a device

Important: These three tools are intentionally restricted to one device at a time. If you ask for software or hardware across all your devices, the AI will ask you to name a specific device instead of looping through all of them. This protects your N-sight API from being overwhelmed on large accounts.

Antivirus

ToolWhat it does
list_av_threatsRecent AV threats on a device
list_av_scansAV scans run on a device
list_av_quarantineItems in AV quarantine on a device

Backup

ToolWhat it does
list_backup_historyBackup check status over the last 90 days
list_backup_sessionsIndividual backup session details

Performance and History

ToolWhat it does
list_performance_historyCPU, memory, and bandwidth metrics over time
list_drive_historyHistorical disk usage over time
list_outagesOutages for a device (open or recently closed)

Other

ToolWhat it does
list_ad_usersActive Directory users at a site
list_client_license_countSoftware license counts by client

Available Tools — Production Server

The production server includes all read-only tools plus write tools for taking action. Use with care — these tools make changes to your N-sight environment.

Additional Write Tools

ToolWhat it does
add_clientCreate a new client
add_siteCreate a new site under a client
clear_checkAcknowledge and clear a failing check
add_check_noteAdd a note to a monitoring check
approve_patchApprove a pending patch for installation
ignore_patchMark a patch as ignored
retry_patchRetry a failed patch installation
run_taskRun a scheduled automation task on a device
start_av_scanTrigger an on-demand AV scan
cancel_av_scanCancel an in-progress AV scan
update_av_definitionsForce an AV definition update
release_quarantine_itemRestore a quarantined file
remove_quarantine_itemPermanently delete a quarantined file

Testing Your Connection

The repo includes an interactive terminal client and an automated test suite.

Interactive terminal (see live logs and notifications)

cd "C:\path\to\N-sight-MCP-Server"
node mcp-terminal.mjs

You'll get a prompt:

nsight>

Common commands:

CommandExample
clientsclients
sites <clientid>sites 129052
devices <siteid>devices 193393
summary <customer name>summary Kelltic Cider Company
hardware <device_name> <assetid>hardware CSP-0009 1771706
software <device_name> <assetid>software CSP-0009 1771706
failingfailing
toolstools
helphelp

This client shows MCP log and progress notifications in real time as the server executes each tool call — useful for confirming the server is working correctly.

Automated test suite

node test-commands.mjs

Runs all 23 read-only tools against your live environment and reports pass/fail for each. Expect output like:

[1/23] tools ......................... PASS (23 tools loaded)
[2/23] clients ....................... PASS (6 clients)
...
[23/23] call list_client_license_count  PASS
All tests passed: 23/23

Guardrails and Rate Limiting

The server includes built-in protections designed for MSPs managing large accounts.

Single-device enforcement. list_software, list_hardware, and list_device_asset_details use an in-memory session guard that blocks repeated calls within a 30-second window. If the AI starts looping through devices, the guard fires and tells it to stop and ask you for a specific device name.

Scope requirements. list_devices and get_environment_summary require context (customer name or site name) before executing. An unscoped call is rejected with a clear error message returned to the AI, which will then prompt you for that context.

Single-customer enforcement. get_environment_summary accepts multiple sites but only within a single customer. Any attempt to span multiple customers is rejected at runtime.

No list_all_devices. This tool was intentionally removed. An account with thousands of devices would generate an enormous API call with no practical value for most queries. Use list_devices scoped to a site instead.


Keeping the Server Up to Date

When you pull new changes from the repo, always rebuild before restarting your AI client:

cd "C:\path\to\N-sight-MCP-Server"
git pull
npm install
npm run build

Then restart Claude Desktop (or your chosen client) to load the updated tools.


Troubleshooting

The AI says it doesn't have any N-sight tools. Check that your config file is valid JSON and the path to readonly-server.js is correct and uses double backslashes on Windows. Restart the AI client after any config change.

The AI asks me for a customer or site name even for basic queries. That's expected behavior for scoped tools. Provide the name and it will proceed.

The AI refuses to list software across all my devices. That's intentional. Name a specific device and it will run the query.

The server starts but returns no clients. Verify your NSIGHT_API_KEY and NSIGHT_SERVER_URL in the .env file. Run node mcp-terminal.mjs and type clients to see the raw response.

Build errors after pulling an update. Run npm install before npm run build to pick up any new dependencies.


Security Notes

Your API key lives only in the .env file in the project directory. It is gitignored and will not be committed if you fork or contribute to the repo. Never paste your API key into your AI client's config file or into a chat session.

The read-only server cannot make changes to your N-sight environment. If you only need monitoring and reporting capabilities, use readonly-server.js exclusively.

The production server has write access. Treat it the same way you would treat direct API access — be specific in your requests and confirm before asking it to take action on devices.

Ver o original no GitHub

Bom para

Relacionados