MCP server exposing the OPNsense API as tools. Speaks JSON-RPC 2.0 over stdio, wraps opnsense.py REST API client.
Server has well-structured tool definitions with clear naming (verb_noun pattern) and comprehensive descriptions (avg 180 chars). All 6 tools have input schemas with type definitions. However, parameter descriptions are sparse or missing in several tools (opnsense_get, opnsense_post lack 'args' descriptions), and output schemas are entirely undocumented. Error handling is present but generic. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Composition is sound, tools chain well via path/controller naming.
Download the running config.xml. Do this before any risky change.
List every command on one controller (e.g. 'firewall/filter') plus the endpoint that commits changes for it. Works without credentials.
Search the offline index of 2399 OPNsense API endpoints by keyword. ALWAYS use this before calling opnsense_get or opnsense_post — endpoint names are not guessable (port forwards live at firewall/d_nat, not firewall/dnat). Works without credentials.
GET an OPNsense API endpoint — read-only. Path is module/controller/command, e.g. 'firewall/filter/search_rule'. Extra URL segments go in args.
POST an OPNsense API endpoint. MUTATES the firewall. A write only edits config.xml — it is not live until you POST the controller's apply or reconfigure endpoint, which this tool names in its response. Use dry_run first to preview. Back up before changing filter rules on a remote box.
Output schemas completely undocumented. No tool returns a documented schema, LLMs cannot predict response structure or plan downstream calls. opnsense_get, opnsense_post, opnsense_probe return JSON but structure is opaque.
Parameter 'args' in opnsense_get and opnsense_post lacks description. Array of strings is ambiguous, LLMs cannot infer whether these are URL path segments, query params, or filter expressions.
No tool annotations despite clear risk profiles. opnsense_post is destructive (WRITE risk) but lacks destructiveHint. opnsense_get, opnsense_find_endpoint, opnsense_describe_controller, opnsense_probe are read-only but lack readOnlyHint. Annotations guide agent safety reasoning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Check connectivity, authentication, firmware version and interfaces.
opnsense_probe description is vague ('Check connectivity, authentication, firmware version and interfaces'). Does not explain what the response contains, when to call it, or what fields are returned.
Error handling is present (opn.ApiError raised) but responses are free-text strings. No structured error classification (retryable vs user-fixable vs fatal). LLMs cannot programmatically determine recovery strategy.