MCP-MCSTATUS

MCP server (Python) with tools for https://mcstatus.xyz/api and Kuma status-page API.

Implemented MCP Tools

• get_minecraft_status - Minecraft server status (Java/Bedrock), endpoint /api/status

• get_java_status - Java status shortcut, endpoint /api/status

• get_bedrock_status - Bedrock status shortcut, endpoint /api/status

• get_srv_records - SRV records, endpoint /api/srv

• resolve_dns - DNS resolution and provider info, endpoint /api/dns

• rdns - reverse DNS (PTR) lookup for IP

• geoip_maxmind - GeoIP lookup using local MaxMind GeoLite2 database

• get_ip_provider_info - provider/operator info for IP via bgp.tools whois + ASN database

• is_ip_anycast - check if player IP is Anycast by curated known-node list

• get_bgp_info - BGP/ASN details for an IP, endpoint /api/bgp

• check_node_status - find Kuma node by name or short alias (e.g., s3, br4) and return UP/DOWN/PENDING/MAINTENANCE

check_node_status For GPT

Use this tool when you need node state from Kuma by human-friendly alias.

Input parameters:

• node_name (string, required) - full node name or short alias.

• timeout_ms (integer, optional, default 4000, must be > 0).

Supported alias patterns:

• full name: s3.joinserver.xyz

• short hostname before first dot: s3 for s3.joinserver.xyz

• token from short name split by -, _, or space: br4

• case-insensitive variants: BR4

• normalized alias (non-alphanumeric chars ignored in fallback matching)

• interpreted noisy aliases with generic prefixes: Node-x21, node x 21, Nodex21, Нода-x21 -> x21.oinserver.xyz

• plain separated aliases also work: x 21 -> x21.oinserver.xyz

Status mapping:

• 1 -> UP

• 0 -> DOWN

• 2 -> PENDING

• any other/unknown -> MAINTENANCE

Result format (ok = true):

{
  "ok": true,
  "input_node_name": "x 21",
  "interpreted_query": {
    "normalized": "x21",
    "core_parts": ["x", "21"],
    "core_fingerprint": "x21"
  },
  "node_name": "x21.oinserver.xyz",
  "node_id": 386,
  "matched_by": "short_hostname_normalized",
  "match_priority": 3,
  "status": "DOWN",
  "status_code": 0,
  "heartbeat_time": "2026-03-15 13:52:53",
  "message": "",
  "ping": null,
  "has_heartbeat": true,
  "matched_by_case_insensitive_name": true
}

Result format (ok = false):

• not found:

{
  "ok": false,
  "input_node_name": "unknown",
  "interpreted_query": {
    "normalized": "unknown",
    "core_parts": ["unknown"],
    "core_fingerprint": "unknown"
  },
  "error": "Node with this name/alias was not found on Kuma status page."
}

• broad or ambiguous alias:

{
  "ok": true,
  "input_node_name": "fra",
  "interpreted_query": {
    "normalized": "fra",
    "core_parts": ["fra"],
    "core_fingerprint": "fra"
  },
  "ambiguous": true,
  "match_count": 2,
  "match_priority": 6,
  "matched_by_modes": ["core_terms_subset"],
  "matches": [
    {
      "node_name": "fra9.joinserver.xyz",
      "node_id": 245,
      "matched_by": "core_terms_subset",
      "match_priority": 6,
      "status": "UP",
      "status_code": 1
    },
    {
      "node_name": "MySQL-FRA9",
      "node_id": 290,
      "matched_by": "core_terms_subset",
      "match_priority": 6,
      "status": "UP",
      "status_code": 1
    }
  ]
}

GPT usage flow:

1. Try the best human string you have: s3, br4, fra28, x 21, Node-x21.

2. If response contains matches, inspect returned statuses directly; broad queries now return all best matches.

3. Only retry with a more specific name when you need a single exact node.

Compatibility note:

• MCP tool responses are returned as regular JSON payloads so FastMCP emits both text content and structuredContent, which avoids "returned no result" behavior in clients that ignore empty structured-only replies.

Architecture

• mcstatus_mcp/client.py - typed API client (MCStatusApiClient)

• mcstatus_mcp/tools.py - abstract BaseMCStatusTool + one class per MCP tool

• mcstatus_mcp/server.py - MCP app bootstrap and tool registration

Each tool is implemented as a class that inherits from BaseMCStatusTool. All tools depend on a shared MCStatusApiClient instance.

Install

.\.venv\Scripts\python.exe -m pip install -r requirements.txt

Run With Docker Compose

docker compose up -d --build

Stop:

docker compose down

Default MCP endpoint from compose:

http://localhost:8000/mcp

Run MCP Server

Default transport is stdio:

.\.venv\Scripts\python.exe main.py

Optional transport override:

$env:MCP_TRANSPORT="sse"
.\.venv\Scripts\python.exe main.py

Allowed MCP_TRANSPORT values: stdio, sse, streamable-http.

Local Stdio Config Example

{
  "mcpServers": {
    "mcstatus": {
      "command": "C:\\Users\\rakse\\PycharmProjects\\MCP-MCSTATUS\\.venv\\Scripts\\python.exe",
      "args": ["C:\\Users\\rakse\\PycharmProjects\\MCP-MCSTATUS\\main.py"]
    }
  }
}

Environment Variables

• MCP_TRANSPORT - MCP transport (default: stdio)

• MCSTATUS_API_BASE_URL - API base URL (default: https://mcstatus.xyz/api)

• KUMA_API_BASE_URL - Kuma API base URL for check_node_status (default: http://status.dsts.cloud:3001/api)

• MCSTATUS_TIMEOUT_MS - default timeout in milliseconds for tools (default: 4000)

• MCP_HOST - host for HTTP transports (sse and streamable-http, default: 127.0.0.1)

• MCP_PORT - port for HTTP transports (default: 8000)

• MCP_STREAMABLE_HTTP_PATH - streamable HTTP path (default: /mcp)

• MCP_SSE_PATH - SSE path (default: /sse)

• MAXMIND_LICENSE_KEY - MaxMind license key for GeoLite2 download (required for auto-download if DB is missing/outdated)

• MAXMIND_DB_PATH - local path to .mmdb file (default: data/GeoLite2-City.mmdb)

• MAXMIND_EDITION_ID - MaxMind edition ID (default: GeoLite2-City)

• MAXMIND_REFRESH_HOURS - database refresh interval in hours; 0 disables periodic refresh (default: 24)

• BGPTOOLS_USER_AGENT - descriptive user-agent with contact for downloading https://bgp.tools/asns.csv (recommended)

• BGPTOOLS_ASN_DB_URL - ASN CSV source URL (default: https://bgp.tools/asns.csv)

• BGPTOOLS_ASN_DB_PATH - local path to ASN CSV cache (default: data/bgp_tools_asns.csv)

• BGPTOOLS_ASN_REFRESH_HOURS - ASN CSV refresh interval in hours; 0 disables periodic refresh (default: 24)

• BGPTOOLS_WHOIS_HOST - bgp.tools whois host (default: bgp.tools)

• BGPTOOLS_WHOIS_PORT - bgp.tools whois port (default: 43)

Use With OpenAI

Important: OpenAI MCP integration uses remote MCP servers. Local stdio servers are good for local dev/testing, but for OpenAI binding you should expose a public HTTPS endpoint.

1) Run as streamable HTTP for deployment

$env:MCP_TRANSPORT="streamable-http"
$env:MCP_HOST="0.0.0.0"
$env:MCP_PORT="8000"
.\.venv\Scripts\python.exe main.py

After deployment your MCP URL will look like:

https://your-domain.example/mcp

2) Bind MCP server in OpenAI Responses API

from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-4.1",
    tools=[
        {
            "type": "mcp",
            "server_label": "mcstatus",
            "server_url": "https://your-domain.example/mcp",
            "require_approval": "never",
            "allowed_tools": [
                "get_minecraft_status",
                "get_java_status",
                "get_bedrock_status",
                "get_srv_records",
                "resolve_dns",
                "rdns",
                "geoip_maxmind",
                "get_ip_provider_info",
                "is_ip_anycast",
                "get_bgp_info",
                "check_node_status",
            ],
        }
    ],
    input="Check DNS and Java status for mc.hypixel.net",
)

print(response.output_text)

3) Bind in ChatGPT (Connectors)

• Open ChatGPT connector settings.

• Add a custom connector using your remote MCP URL.

• Select tools and permissions.

Reference docs:

• https://platform.openai.com/docs/guides/tools-remote-mcp

• https://platform.openai.com/docs/guides/mcp

• https://help.openai.com/en/articles/11487775-connectors-in-chatgpt/