clearon-ops-audit-mcp
# Clearon Ops Audit MCP
Read-only MCP server for deterministic scheduler, runtime, and lock-state audits.
It summarizes recurring-job health, delivery failures, degraded-success patterns, overdue runs, gateway/MCP health, and lock-file inventory without modifying the system it inspects.
## What it does
- Lists recurring jobs and summarizes current job health
- Separates execution failures from delivery failures
- Flags repeated degraded-success summaries
- Reports overdue recurring jobs
- Probes gateway, MCP, and approved lock roots in read-only mode
- Generates a combined operational report
## What it does not do
- It does not edit cron jobs or configuration
- It does not restart services
- It does not delete lock files
- It does not infer that a lock file is stale just because it exists
## Requirements
- Node.js `20` or newer
- The `openclaw` CLI available on `PATH`, or `OPENCLAW_BIN` set explicitly
- A local MCP client that can launch a stdio server
## Installation
```sh
cd tools/clearon-ops-audit-mcp
npm install
npm test
npm run smoke
```
## MCP client setup
Example stdio configuration:
```json
{
"mcpServers": {
"clearon-ops-audit": {
"command": "node",
"args": ["/absolute/path/to/clearon-ops-audit-mcp/server.mjs"],
"env": {
"CLEARON_OPS_AUDIT_WORKSPACE": "/absolute/path/to/workspace"
}
}
}
}
```
The workspace defaults to the client process working directory. Override it with `CLEARON_OPS_AUDIT_WORKSPACE` when lock scanning should target a specific tree.
## Commands
```sh
npm test
npm run smoke
npm run syntax
npm run report
```
## Tools
- `get_server_health`
- `list_recurring_jobs`
- `get_job_health`
- `get_recent_failures`
- `find_missed_or_overdue_runs`
- `summarize_degraded_successes`
- `check_delivery_failures`
- `probe_registered_mcps`
- `check_gateway_health`
- `audit_locks`
- `generate_ops_report`
TDQS
Scored across 11 tools
Several tools cluster around health and failure detection—get_server_health/check_gateway_health/probe_registered_mcps overlap in probing MCP/gateway state, and get_recent_failures/check_delivery_failures/find_missed_or_overdue_runs all surface different failure modes. Descriptions clarify the distinctions, but an agent could misroute a health/failure query without careful reading.
All tool names are snake_case and mostly follow a verb_noun pattern, but verb choices are inconsistent: get/list/find/summarize/probe/check/audit/generate are used interchangeably, and check_/get_ overlap semantically. Still readable and object-focused.
11 tools is well within the sweet spot for a specialized read-only ops audit server. Each tool corresponds to a distinct audit concern (health, jobs, failures, MCPs, delivery, locks, reporting) without bloat.
The surface covers the main read-only audit workflow: enumerate jobs, assess job/server/gateway health, surface failures and anomalies, audit locks, and generate a report. Minor gaps like per-tool historical trend endpoints or more granular failure detail are workaround-able.