MCP server
What is the AdminKit MCP server?
AdminKit's Model Context Protocol server lets an AI agent call AdminKit's checks directly, instead of
guessing at a user's IP, DNS or MCP exposure. Open source —
github.com/netbasecorp/adminkit-mcp,
MIT licensed, version 0.1.0 — published on
PyPI as adminkit-mcp and
listed in the MCP Registry.
The hosted endpoint AdminKit plans to run is not live yet.
Updated
Tools
check_ip Look up the caller's own IP, ASN, approximate location and CGNAT hint.
dns_lookup Raw DNS record lookup for a name and record type.
check_mail_auth SPF and DMARC validation for a domain.
check_domain Aggregate whois, TLS and HTTP-header overview for a domain.
check_mcp_endpoint Probe an MCP server for anonymous access and auth/TLS posture.
check_owned_ports Open-port check limited to a target you've verified you own.
explain_report AI-generated, citation-backed explanation of a saved check report.
get_radar Aggregated honeypot/telemetry exposure trend data (k ≥ 5 only).
subnet_calc IPv4/IPv6 subnet details for a CIDR: network, usable range, flags.
ip_info Inspect any IP address: version, private/global/loopback flags, notation.
ipv6_info Inspect an IPv6 address: /64 network, reverse-pointer form, embedded IPv4.
mac_vendor Look up the registered vendor for a MAC address or OUI prefix.
user_agent Parse a User-Agent string into browser, OS, device family and bot flag.
Free connector
The hosted endpoint above is not live yet, but the free connector behind it is real, tested, public code you can run yourself right now:
uvx adminkit-mcp
It is published on PyPI as
adminkit-mcp 0.1.0 (also installable with
pip install adminkit-mcp) and listed in the MCP Registry as
io.github.netbasecorp/adminkit-mcp. Prefer installing straight from the git tag instead? Run
uvx --from git+https://github.com/netbasecorp/[email protected] adminkit-mcp. By default it calls AdminKit's Check API (api.adminkit.net),
which is live in production — so all 13 tools return real data today, with a resilience fallback for three of
them and an honest "not available" result instead of fabricated data if a route is ever briefly unreachable.
See connect an AI agent to AdminKit's MCP server for
the full setup and per-tool breakdown.
check_mcp_endpoint Always works — tries the live Check API first, falls back to a local, self-contained probe either way.
check_ip Calls the live Check API; falls back to AdminKit's live site function if that route is ever unreachable.
dns_lookup Calls the live Check API; falls back to the live site function if that route is ever unreachable.
check_mail_auth Calls the live Check API; falls back to the live site function if that route is ever unreachable.
check_domain, check_owned_ports, get_radar Call the live Check API directly (no site fallback); report a clear "not available" result instead of fabricating one if a route is briefly unreachable.
explain_report Fetches a live, saved report and asks the live AI explainer — needs your own ADMINKIT_API_KEY, or it reports "not available."
subnet_calc, ip_info, ipv6_info, mac_vendor, user_agent Pure local computation, no network call at all — always available.
The same probe behind check_mcp_endpoint also
ships as a GitHub Action connector
for running the same check in CI, on every push or a schedule.
Example request (coming with the API)
This endpoint is not live yet, but the request and response shapes below are not invented — they follow the
Model Context Protocol's JSON-RPC 2.0 message format (specification version 2025-06-18) exactly, so
integrators can build against it ahead of launch. Point an MCP client at https://mcp.adminkit.net/mcp once
it ships.
1. Initialize the session
curl -s https://mcp.adminkit.net/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "initialize",
"params": {
"protocolVersion": "2025-06-18",
"capabilities": {},
"clientInfo": { "name": "example-client", "version": "1.0.0" }
}
}' Expected response shape:
{
"jsonrpc": "2.0",
"id": 1,
"result": {
"protocolVersion": "2025-06-18",
"capabilities": { "tools": {} },
"serverInfo": { "name": "adminkit-mcp", "version": "0.1.0" }
}
} 2. List available tools
curl -s https://mcp.adminkit.net/mcp \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{"jsonrpc": "2.0", "id": 2, "method": "tools/list"}' Expected response shape (trimmed to two tools):
{
"jsonrpc": "2.0",
"id": 2,
"result": {
"tools": [
{ "name": "check_ip", "description": "Look up IP, ASN, approximate location and CGNAT hint for an address." },
{ "name": "check_mcp_endpoint", "description": "Check whether an MCP server requires authentication." }
]
}
} Frequently asked questions
Is the MCP server live yet?
The hosted endpoint planned at mcp.adminkit.net/mcp is not live yet — it still needs its DNS route and tunnel brought up on AdminKit's side. But the open-source connector you self-host is: the repository (https://github.com/netbasecorp/adminkit-mcp, tag v0.1.0, MIT) is public, and it talks to AdminKit's Check API (api.adminkit.net), which went live in production on 01/11/2026 — see "Tools" below for what that means for each tool.
Will the MCP server require authentication?
The self-hosted connector works without an API key for every tool except explain_report, which needs your own ADMINKIT_API_KEY for the AI explanation step. The future hosted endpoint is expected to follow the same pattern: free-tier calls without an account, within a rate limit, and an API key for higher-volume or account-scoped calls.
Where will the MCP server be hosted?
At mcp.adminkit.net/mcp, a public endpoint maintained by AdminKit (Netbase JSC) — not live yet (see /contact/ for updates). The server's own source code is already public and MIT-licensed, independent of when the hosted endpoint ships: see /legal/terms/ for how that differs from AdminKit's own paid plans.
Can I run the MCP server myself before the hosted endpoint ships?
Yes. Install it with `uvx adminkit-mcp` or `pip install adminkit-mcp` — the package is published on PyPI as `adminkit-mcp` 0.1.0 and listed in the MCP Registry as `io.github.netbasecorp/adminkit-mcp`. All 13 tools call a real backend; see our self-hosting guide for the full per-tool breakdown.
What happens if the Check API is briefly unreachable while a tool is running?
It depends on the tool: check_mcp_endpoint, check_ip, dns_lookup and check_mail_auth fall back to a local, self-contained probe or AdminKit's live site function, so they keep working. check_domain, check_owned_ports and get_radar call the Check API directly with no fallback — they report an honest "not available" result instead of guessing or returning fabricated data. subnet_calc, ip_info, ipv6_info, mac_vendor and user_agent never touch the network at all, so an API outage never affects them.
Install the MCP server today
The repository is public and MIT licensed — there is no mailing list yet for the hosted endpoint, but contact us and we will tell you when it ships.