The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Netmon listing page.
Connect an AI assistant to a Netmon appliance.
Netmon ships an MCP server on the appliance itself. This repository holds the client-side bridge: a small Node process that runs on your own machine, speaks stdio to Claude Desktop, and forwards every JSON-RPC message to your Netmon's /mcp endpoint over HTTPS.
Once connected, an assistant can read your devices, interfaces, logs, flow records, alerts, and Windows-agent data by calling Netmon's tools directly — with the same sign-in and permissions as any operator, and the same tag restrictions.
A public, read-only demo server answers the same 36 tools from a recorded, de-identified snapshot of a fictional 90-device network:
No account, no token, no install. Point any MCP client at it:
GUI clients — Claude Desktop (Settings → Connectors → Add custom connector) and ChatGPT (developer mode → Connectors) — take the URL with authentication set to None. For Cursor, Zed, Kimi Code CLI, Continue and LM Studio, drop the URL into that client's MCP config file:
Then ask it something real: "what's down?", "show me the Suricata alerts from last week", "which host is using the most bandwidth, and who is it talking to?"
It is a fixture, not an appliance: nothing writes, nothing reaches a real network, and every answer is the same shape a live Netmon returns — with timestamps shifted forward so the snapshot always reads as the last day or so.
Two separate walls stand between an MCP client and an appliance on your LAN, and the bridge exists for both.
Where the connection comes from. Claude Desktop's remote-connector option and ChatGPT's custom connectors both fetch the MCP URL from their vendor's cloud, not from your machine — Anthropic's docs say so outright, and their connector resolves the hostname and rejects private addresses before any request leaves their network, so your appliance's log stays empty. Most other clients (Claude Code, Codex, Gemini CLI, Cursor, VS Code, Zed, Cline, Continue, opencode, Goose, LM Studio, Kimi) connect from your own machine and are unaffected.
The certificate. This one catches nearly everyone. Most clients validate TLS against the operating system trust store and expose no skip-verify option, and an appliance ships a self-signed certificate — so the route succeeds and the handshake fails. Either install the appliance's certificate in the machine's trust store, or use the bridge, which pins it on first use.
The bridge runs as a local process and handles both. It is not a Claude-specific workaround; it is the general answer.
If your appliance is internet-facing with a publicly-trusted certificate, you can skip the bridge and point any MCP client straight at the endpoint. See Install — any MCP client below.
mcp:* scopes you want the assistant to have..mcpb install. Claude Desktop supplies its own Node runtime, so nothing else is needed. Running the bridge outside Claude Desktop requires Node 18 or newer.Download netmon-mcp-<version>.mcpb from Releases. The same bundle ships with your appliance — Settings → System → Downloadables → Claude Desktop Extension.
In Claude Desktop, open Settings → Extensions → Advanced settings → Extension Developer → Install Extension… and select the file.
Fill in the two settings the extension asks for:
| Setting | Value |
|---|---|
| Netmon URL | https://<your-netmon>/mcp — the full endpoint, including /mcp |
| API Token | a personal access token (see below). The Bearer prefix is added for you if you leave it off. |
Restart the extension. Ask Claude something like "list the devices that are down" to confirm it is working.
In Netmon, go to Settings → Users, click the key icon on your user, and create a personal access token. Select only the scopes the assistant needs — the token cannot be widened later, and it can be revoked from the same screen at any time.
The account must hold the API permission. Without it, token minting and OAuth consent are both refused.
The appliance serves Streamable HTTP at https://<your-netmon>/mcp (POST for JSON-RPC, GET for the server-to-client SSE stream, DELETE to end a session).
Clients that support OAuth 2.1 discover the authorization server on their own: the endpoint answers an unauthenticated request with 401 and a WWW-Authenticate: Bearer resource_metadata="…" challenge pointing at RFC 9728 protected-resource metadata, alongside RFC 8414 authorization-server metadata and dynamic client registration at /auth/register. You sign in through your browser on your own Netmon and approve the scopes on a consent page.
Clients without OAuth support pass a token directly:
Most appliances serve a self-signed certificate. A direct type: http entry will fail against one — the OAuth bootstrap rejects the certificate and does not consult NODE_EXTRA_CA_CERTS. Use the bridge instead, which handles the certificate itself:
36 read tools, each wrapping a Netmon API endpoint and gated by an OAuth scope that is checked before the call runs.
| Area | Scope | Example tools |
|---|---|---|
| Devices & fleet | mcp:devices | device_find, device_get, device_list, device_metric_summary, overwatch_summary, tags_list |
| Traffic & topology | mcp:vne, mcp:devices | top_bandwidth, netflow_search, netflow_raw_search, flow_summary, get_network_entity_info, arp_table, interfaces_search |
| Logs & security | mcp:logs | syslog_search, eventlog_search, eve_search, eve_get, log_severity_summary, syslog_facets |
| Alerts | mcp:alerts | alerts_list, alerts_history, maintenance_windows_list |
| Windows agent | mcp:devices | agent_processes, agent_services, agent_disk_usage |
| Live tools | mcp:tools, mcp:devices, mcp:capture | ping, traceroute, arp_lookup, port_map, snmp_test, snmp_walk_run, snmp_walk_last, search_ip, speedtest_history, capture_list, capture_get |
| Tool | Title | What it answers |
|---|---|---|
get_network_entity_info | Look Up WHOIS, GeoIP and DNS | Retrieves WHOIS, GeoIP and DNS information for a public IP address or hostname. |
arp_lookup | ARP Lookup | Performs an ARP lookup to find the MAC address for a given Local IP address. |
ping | Ping Host | Ping a target host from the Netmon server. |
traceroute | Run Traceroute | Traceroute to a target from the Netmon server. |
port_map | Scan Open Ports | Nmap port scan against a single host from the Netmon server. |
search_ip | Search All Streams for an IP | Find every mention of a specific IP across Netmon's log and telemetry streams: syslog, Windows eventlog, Suricata EVE, aggregated NetFlow, and ARP. |
syslog_search | Search Syslog | Search syslog messages from network devices. |
eventlog_search | Search Windows Event Log | Search Windows Event Log entries ingested from Netmon agents. |
eve_search | Search Suricata IDS Events | Search Suricata EVE-format IDS events. |
eve_get | Get Suricata IDS Event | Fetch a single Suricata EVE event by id, decoded server-side. |
log_severity_summary | Summarize Log Severities | Count log events grouped by severity over a time window. |
syslog_facets | Syslog Facet Breakdown | Top-N value counts for ONE syslog field over a window — 'what are the top actions/reasons on this FortiGate in the last 2 hours' in a single call,... |
netflow_search | Search Aggregated NetFlow | Search the FULL NetFlow history: the raw flow table (the last ~15 minutes) unioned with the aggregated rollup (4 weeks of history), windowed and pr... |
netflow_raw_search | Search Raw NetFlow | Search raw NetFlow records (per-flow, not aggregated). |
flow_summary | Summarize NetFlow Traffic | Summarize one host's network conversations: top peers, top ports, and a client-vs-service-side split, each with a residual "other" bucket plus over... |
device_list | List Devices | List monitored devices. |
device_get | Get Device Details | Fetch one device with its related state: tags, alerts, the ping / oid / interface / port / disk trackers configured on it, its SNMP walk trackers,... |
device_find | Find Devices by Name or IP | Find devices matching a substring of label or ip_address. |
overwatch_summary | Overwatch Health Summary | High-level network health snapshot for 'how's the network?' style questions. |
snmp_test | Test SNMP Credentials | Probe a device for SNMP reachability using the Netmon snmptest binary. |
snmp_walk_last | Get Cached SNMP Walk | Fetch the most recent stored SNMP walk for a device (cached in tools_walks). |
snmp_walk_run | Run Fresh SNMP Walk | Trigger a FRESH SNMP walk against a device. |
agent_services | List Windows Services | List Windows services on an agent-managed device — live read via the WMI tunnel. |
agent_processes | List Windows Processes | List running processes on an agent-managed device — live read via the agent tunnel. |
alerts_list | List Alert Definitions | List configured alert definitions across both axes of the rule engine. |
alerts_history | Alert Incident History | Authoritative 'what fired and when' stream — wraps the alert_history table (one row per incident, both legacy and modern) and alert_outlet_log... |
maintenance_windows_list | List Maintenance Windows | Lists maintenance windows — the suppression schedules that gate alert dispatch. |
tags_list | List Tags | List tag definitions. |
top_bandwidth | Top Bandwidth Talkers | Top NetFlow conversations over the last N minutes — the 'who's eating bandwidth right now?' question. |
arp_table | Get ARP Table | Lists hosts observed on the local LAN(s) via the ARP table — the 'what devices have we seen recently?' question. |
interfaces_search | Search Network Interfaces | Cross-device interface metadata listing — answers 'what interfaces are tracked across the fleet, named like X, on device Y?'. |
device_metric_summary | Summarize Device Metrics | Day / week / month / all-time summary stats for a single device-tracker, by metric type. |
agent_disk_usage | Browse Agent Disk Usage | Path-scoped folder-tree disk usage report from a Netmon agent — the 'D: drive is at 95%, what's eating it?' question. |
capture_list | List Packet Captures | Read-only listing of packet captures. |
capture_get | Get Packet Capture Status | Read-only single-capture detail. |
speedtest_history | Speedtest History | Recent WAN speedtest results — answers 'is the internet healthy?'. |
Eight scopes exist in total — the six above plus mcp:reports and mcp:system, which map to their permissions but have no tools yet.
Everything stays on your appliance. The bridge connects to the one URL you configure and nothing else. No telemetry, no vendor endpoint, no third-party service sits in the path. Netmon Services receives nothing.
Read-only. Every tool is a read. Netmon's agent write paths — killing a process, starting or stopping a Windows service, deleting a file, running PowerShell — are permanently excluded at the MCP layer, regardless of what the underlying token's permissions would otherwise allow. An assistant cannot change your monitoring configuration, acknowledge an alert, or act on a managed host.
Scoped. Each tool declares the OAuth scope it requires, and the scope is checked before the tool body runs. A token granted mcp:logs only cannot call device_list.
Yours, and no wider. A token acts as the user who minted it. Tag restrictions apply: an operator limited to the branch-offices tag gets an assistant limited to the same devices. The permission model is the one already governing the web interface and every other API client.
Revocable. Revoke a token from Settings → Users and the assistant loses access immediately. Tokens can also be scoped narrowly at mint time rather than revoked later.
On first connection the bridge captures the appliance's TLS certificate and stores it at ~/.netmon-mcp/<hash-of-url>.pem, then verifies every later connection against that pinned certificate. This is trust-on-first-use: it protects against interception after the first connection, not during it. Make the first connection from a network you trust.
If you replace the appliance's certificate, delete the stale pin (rm ~/.netmon-mcp/*.pem) or the bridge will verify the new certificate against the old one and refuse to connect.
The bridge requires https://. Plain HTTP is rejected at startup.
The bridge writes a diagnostic log to ~/.netmon-mcp/wrapper.log.
| Symptom | Cause |
|---|---|
NETMON_URL must use https:// | The URL is http://, or the scheme is missing. |
TLS probe failed | The appliance is unreachable on 443 from this machine, or the hostname does not resolve. |
| Connects, then fails after a certificate change | Stale pin. Delete ~/.netmon-mcp/*.pem and reconnect. |
| Tools return an authorization error | The token lacks the scope that tool requires, or the account lacks the API permission. |
| No tools appear | The URL is missing the /mcp path. |
Output lands in build/netmon-mcp-<version>.mcpb.
Full policy: https://netmon.com/privacy-policy/
What is collected. This bridge collects no data. It holds no account, no analytics, and no callback to Netmon Services. Two files are written locally, both under ~/.netmon-mcp/: the pinned certificate of the appliance you configured, and a diagnostic log of connection events. Your Netmon URL and token are supplied by you and are held only in the extension's own configuration.
How data is used and stored. Queries and results pass between your MCP client and your appliance. The bridge forwards messages and retains none of them. All monitoring data stays in the appliance's database, under your control, in your building.
Third-party sharing. None. The bridge contacts exactly one host — the appliance URL you configure. It sends nothing to Netmon Services or any other party.
Note that your MCP client is a separate product with its own policy. When you ask an assistant a question, the tool results it receives are handled by that client's vendor under their terms. Netmon does not control that leg, which is why the scopes on a token are worth setting narrowly.
Retention. Local files persist until you delete them: remove the extension, or rm -rf ~/.netmon-mcp. Revoking the token in Settings → Users ends access immediately. Retention of the monitoring data itself is configured on your appliance, by you.
Contact. support@netmon.com
Mozilla Public License 2.0 — see LICENSE.