mcp-nagios-crunchtools
# mcp-nagios-crunchtools
Secure MCP server for Nagios Core monitoring. Query host and service status, acknowledge problems, add comments, schedule forced checks, and read notification history.
## Installation
```bash
uvx mcp-nagios-crunchtools
```
## Configuration
| Variable | Required | Description |
|----------|----------|-------------|
| `NAGIOS_URL` | Yes | Base URL (e.g., `https://nagios.example.com`) |
| `NAGIOS_USER` | Yes | HTTP Basic Auth username |
| `NAGIOS_PASS` | Yes | HTTP Basic Auth password |
| `NAGIOS_COMMENT_AUTHOR` | No | Author recorded on acknowledgements and comments. Defaults to `NAGIOS_USER`. Not a tool parameter. |
| `NAGIOS_CGI_PREFIX` | No | Path under `NAGIOS_URL` where the CGI scripts are served. Defaults to `/nagios/cgi-bin`. Set to `/cgi-bin` for a Nagios served at the web root. A trailing slash is ignored. |
| `NAGIOS_LOG_PATH` | No | Read-only path to `nagios.log`. Enables `nagios_downtime_history` and the planned-work split in `nagios_alert_history`; archivejson.cgi never returns downtime entries. |
| `TZ` | Effectively yes | The container's local timezone. Must match the Nagios server's own timezone. |
### Why `TZ` matters
`schedule_check` submits a `start_time` to `cmd.cgi`, which parses it as naive
wall-clock time in the Nagios server's local timezone — there is no offset field
to send. The value is therefore built with `time.localtime()`, so this container
must run in the same timezone as the Nagios server.
When it does not, forced checks are silently scheduled into the future by the
offset between the two — four hours against an EDT server — and "check now"
appears to do nothing at all. No error is raised. This was a real defect, fixed
in 0.1.2 by switching from `gmtime()` to `localtime()`; the timezone assumption
it introduced is why `TZ` belongs in this table rather than being left implicit.
## Tools
| Tool | Description |
|------|-------------|
| `nagios_host_status` | Query a single host's status |
| `nagios_service_status` | Query a single service's status |
| `nagios_current_problems` | All hosts/services not in OK state, with duration, attempt, acknowledgement and plugin output |
| `nagios_program_status` | Health check on the monitoring path itself (wedged daemon, disabled notifications, stale status) -- not the same as no current problems |
| `nagios_acknowledge` | Acknowledge a host or service problem |
| `nagios_add_comment` | Add a comment to a host or service |
| `nagios_schedule_check` | Force an immediate re-check |
| `nagios_schedule_downtime` | Fixed downtime from now (5–480 min) on a host and its services, or one service, before planned work |
| `nagios_list_downtimes` | Downtime in effect or pending |
| `nagios_cancel_downtime` | End this server's own downtime when the work is done |
| `nagios_downtime_history` | Past and current downtime windows with author and comment, from `nagios.log` |
| `nagios_notification_history` | Recent notifications |
| `nagios_alert_history` | State changes rolled up per host/service, including warnings and problems that recovered by themselves; problems during scheduled downtime are totalled separately |
## License
AGPL-3.0-or-later
TDQS
Scored across 13 tools
Each tool targets a distinct action or data view, and descriptions actively disambiguate tricky pairs (program_status vs current_problems, alert_history vs notification_history, downtime_history vs list_downtimes). The only mild overlap is between downtime_history ('past and current') and list_downtimes ('in effect or still to come'), which could still confuse an agent about which covers active windows.
All 13 tools follow the same nagios_<verb/noun>_tool pattern (nagios_acknowledge_tool, nagios_schedule_downtime_tool, etc.), with no camelCase or style mixing. Predictable and easy to scan.
13 tools is well within the ideal range and each maps to a concrete monitoring operation (status queries, acknowledgements, comments, downtime lifecycle, history, self-health). No filler or redundant tools are apparent.
Core read lifecycle (host/service status, current problems, program status) and downtime lifecycle (schedule/cancel/list/history) plus acknowledgement and comments are covered. Minor gaps exist: no unacknowledge/remove-comment, no enable/disable notifications, and no host/service inventory enumeration, but these are workaroundable.