Dashboard & Monitoring

Overview

MCP Filesystem Ultra includes an audit logging system and a separate web dashboard binary for monitoring operations, backups, normalizer activity, and error patterns.

The system has two parts:

Audit logging — enabled via --log-dir on the MCP server, writes JSON Lines logs and metrics snapshots
Dashboard binary — separate dashboard.exe that reads those log files and serves a web UI

There is no direct coupling between the MCP server and the dashboard. Communication is file-based only.

Audit Logging

Enabling

Add --log-dir to the MCP server args:

{
  "mcpServers": {
    "filesystem-ultra": {
      "command": "C:\\path\\to\\filesystem-ultra-v4.exe",
      "args": [
        "--log-dir", "C:\\Logs\\MCP",
        "C:\\project",
        "C:\\Logs\\MCP"
      ]
    }
  }
}

Note: Include the log directory in allowed paths so the dashboard can read it.

When --log-dir is not set, audit logging is disabled with zero overhead — the auditWrap() middleware returns the handler unchanged.

Files Written

File	Format	Description
`operations.jsonl`	JSON Lines	One entry per tool call. Auto-rotates at 10 MB, keeps last 3 files.
`metrics.json`	JSON	Performance snapshot, updated every 30 seconds.
`normalizer_stats.json`	JSON	Normalizer activity: per-tool counts, per-rule hits, recent normalizations.

Audit Entry Fields

Each line in operations.jsonl contains:

Field	Description
`ts`	Timestamp (ISO 8601)
`tool`	Tool name (e.g., `edit_file`, `search_files`)
`path`	Primary file path
`duration_ms`	Execution time in milliseconds
`bytes_in` / `bytes_out`	Bytes received / returned
`status`	`ok` or `error`
`error`	Error message (when status is `error`)
`risk`	Risk level (`LOW`, `MEDIUM`, `HIGH`, `CRITICAL`)
`file_size`	Size of the primary file
`args`	Summarized parameters
`sub_op`	Sub-operation detail (e.g., `step:2/5:edit-step:edit` for pipeline steps)
`lines_changed`	Number of lines modified
`matches`	Number of search/count matches
`cache_hit`	Whether the operation hit cache
`norms`	Normalizations applied (rule ID, param, from/to)

All 16 tools are wrapped with auditWrap() in main.go.

Metrics Snapshot

metrics.json contains:

{
  "updated_at": "2026-03-16T10:30:00Z",
  "ops_total": 1234,
  "ops_per_sec": 2.5,
  "cache_hit_rate": 0.92,
  "memory_mb": 85.3,
  "reads": 450,
  "writes": 120,
  "lists": 200,
  "searches": 300,
  "edits": {
    "total": 164,
    "targeted": 140,
    "rewrites": 24,
    "avg_bytes": 350.5
  }
}

Pipeline Progress

When a pipeline runs with --log-dir enabled, each completed step emits a separate audit entry with sub_op: "step:N/M:stepID:action". This enables per-step visibility in the dashboard.

Dashboard Binary

Build and Run

# Build
go build -ldflags="-s -w" -trimpath -o dashboard.exe ./cmd/dashboard/

# Run
dashboard.exe --log-dir=C:\Logs\MCP --backup-dir=C:\Backups\MCP --port=9100

Or use the included run-dashboard.bat.

CLI Flags

Flag	Default	Description
`--log-dir`	(required)	Same directory as the MCP server’s `--log-dir`
`--backup-dir`	(required)	Same directory as the MCP server’s `--backup-dir`
`--port`	9100	HTTP port for the web UI

Architecture

Single binary with embedded web assets (go:embed)
Reads operations.jsonl and metrics.json from --log-dir
Real-time updates via Server-Sent Events (SSE)
Backup cache with 30-second TTL (avoids repeated disk scans)
Unified backup format: both normal backups and batch backups normalized into a common structure

Dashboard Pages

Dashboard (Metrics)

Live overview of server activity:

Operations per second
Cache hit rate
Memory usage
Operation breakdown (reads, writes, searches, edits)
Edit telemetry (targeted vs. rewrites, average bytes)

Operations (Audit Log)

Searchable, filterable list of all tool invocations:

Timestamp, tool name, path, duration, status
Risk level indicators
Normalizations applied
Sub-operation detail for pipeline steps

Backups (Recovery)

Enterprise search/filter/recovery system:

Summary cards: Total backups, total size, latest backup, protected files
Search: Text filter (file name, backup ID, context), operation type dropdown, date presets (today, 24h, 7d, 30d, custom range)
Content search: Grep inside backup files with context snippets (2 lines before/after match, 10-second timeout)
Pagination: Server-side with configurable limit/offset

Backup API Endpoints

Endpoint	Description
`GET /api/backups`	All backups (cached, unified format)
`GET /api/backups/search?q=&operation=&preset=&from=&to=&limit=&offset=`	Filtered + paginated
`GET /api/backups/search-content?q=&max_results=`	Grep inside backup files
`GET /api/backups/detail/{id}`	Single backup with file details
`GET /api/backups/file/{id}/{filename}`	Serve backup file content

Statistics

Aggregate statistics over the session:

Operations by tool
Top paths by access count
Error rates
Duration percentiles

Normalizer

Normalizer activity monitoring:

Total requests processed vs. normalized
Per-tool normalization counts
Per-rule hit counts (which rules fire most)
Recent normalizations with timestamps

Edit Analysis

Edit-specific telemetry:

Targeted edits vs. rewrites
Average bytes per edit
Risk level distribution
Edit patterns over time

Error Patterns

Error monitoring and analysis:

Errors by tool
Error messages grouped by pattern
Recent errors with full context

Configuration Example

Full observability setup:

{
  "mcpServers": {
    "filesystem-ultra": {
      "command": "C:\\path\\to\\filesystem-ultra-v4.exe",
      "args": [
        "--compact-mode",
        "--log-dir", "C:\\Logs\\MCP",
        "--log-level", "info",
        "--backup-dir", "C:\\Backups\\MCP",
        "--backup-max-age", "30",
        "--backup-max-count", "500",
        "C:\\project",
        "C:\\Backups\\MCP",
        "C:\\Logs\\MCP"
      ]
    }
  }
}

Then run the dashboard:

dashboard.exe --log-dir=C:\Logs\MCP --backup-dir=C:\Backups\MCP --port=9100

Access at http://localhost:9100.