/ Verzeichnis / Playground / claude-code-organizer
● Community mcpware ⚡ Sofort

claude-code-organizer

von mcpware · mcpware/claude-code-organizer

Audit Claude Code memories, MCPs, and skills in a dashboard — plus a scanner that catches tool-poisoning attempts.

claude-code-organizer by mcpware is a local dashboard + MCP scanner. It shows exactly what Claude Code loads into context (with per-item token counts), flags MCP servers with suspicious or obfuscated tool descriptions, and helps clean duplicate configs across project and global scopes.

Warum nutzen

Hauptfunktionen

Live-Demo

In der Praxis

claude-code-organizer.replay ▶ bereit
0/0

Installieren

Wählen Sie Ihren Client

~/Library/Application Support/Claude/claude_desktop_config.json  · Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "claude-code-organizer": {
      "command": "npx",
      "args": [
        "-y",
        "claude-code-organizer"
      ],
      "_inferred": true
    }
  }
}

Öffne Claude Desktop → Settings → Developer → Edit Config. Nach dem Speichern neu starten.

~/.cursor/mcp.json · .cursor/mcp.json
{
  "mcpServers": {
    "claude-code-organizer": {
      "command": "npx",
      "args": [
        "-y",
        "claude-code-organizer"
      ],
      "_inferred": true
    }
  }
}

Cursor nutzt das gleiche mcpServers-Schema wie Claude Desktop. Projektkonfiguration schlägt die globale.

VS Code → Cline → MCP Servers → Edit
{
  "mcpServers": {
    "claude-code-organizer": {
      "command": "npx",
      "args": [
        "-y",
        "claude-code-organizer"
      ],
      "_inferred": true
    }
  }
}

Klicken Sie auf das MCP-Servers-Symbol in der Cline-Seitenleiste, dann "Edit Configuration".

~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "claude-code-organizer": {
      "command": "npx",
      "args": [
        "-y",
        "claude-code-organizer"
      ],
      "_inferred": true
    }
  }
}

Gleiche Struktur wie Claude Desktop. Windsurf neu starten zum Übernehmen.

~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "claude-code-organizer",
      "command": "npx",
      "args": [
        "-y",
        "claude-code-organizer"
      ]
    }
  ]
}

Continue nutzt ein Array von Serverobjekten statt einer Map.

~/.config/zed/settings.json
{
  "context_servers": {
    "claude-code-organizer": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "claude-code-organizer"
        ]
      }
    }
  }
}

In context_servers hinzufügen. Zed lädt beim Speichern neu.

claude mcp add claude-code-organizer -- npx -y claude-code-organizer

Einzeiler. Prüfen mit claude mcp list. Entfernen mit claude mcp remove.

Anwendungsfälle

Praxisnahe Nutzung: claude-code-organizer

Audit what's preloaded into every Claude Code session

👤 Power users whose context starts half-full before they type ⏱ ~15 min beginner

Wann einsetzen: Responses feel slow or context fills fast and you suspect too many memories/skills.

Voraussetzungen
  • Claude Code installed with existing memories/MCPs — Standard CC setup
Ablauf
  1. Launch the dashboard
    Run npx @mcpware/claude-code-organizer and open the dashboard.✓ Kopiert
    → Browser opens with all scopes visible
  2. Sort by token cost
    In the dashboard, sort memories/skills by token count — what are the top 5 biggest?✓ Kopiert
    → Clear top-offender list
  3. Prune ruthlessly
    Delete or move to project scope the ones that aren't global concerns.✓ Kopiert
    → Token budget drops

Ergebnis: Smaller preloaded context, faster sessions, lower per-call cost.

Fallstricke
  • Deleting a memory that a teammate relies on — If it's team-wide, move to project scope rather than delete

Scan your MCP servers for tool-poisoning attempts

👤 Anyone installing community MCPs ⏱ ~10 min intermediate

Wann einsetzen: Before or after installing an MCP from an unfamiliar publisher.

Ablauf
  1. Run the scanner
    Scan all configured MCP servers for suspicious patterns in tool descriptions.✓ Kopiert
    → Report with NEW/CHANGED/CLEAN per server
  2. Investigate flags
    For every flagged finding, show the exact tool description and which pattern fired.✓ Kopiert
    → Explanations linking pattern → evidence
  3. Disable or remove bad actors
    Disable the flagged MCP from project scope and re-scan.✓ Kopiert
    → Clean second run

Ergebnis: MCPs that try to smuggle instructions via tool descriptions are caught before they compromise a session.

Fallstricke
  • False positives on legitimate servers using unusual language — Review the pattern match; false positives don't hurt, but re-enable after verifying
  • Scanner only catches embedded-prompt-style attacks, not all malicious behavior — Still review source for MCPs that execute code on your box

Deduplicate MCP configs across global and project scopes

👤 Developers with layered .mcp.json + ~/.claude configs ⏱ ~5 min beginner

Wann einsetzen: Same MCP registered twice in different scopes causing odd behavior.

Ablauf
  1. Detect duplicates
    Find MCP servers registered in multiple scopes.✓ Kopiert
    → List of duplicates with their scopes
  2. Consolidate
    Keep the project-scope definition; remove from global.✓ Kopiert
    → Single canonical entry

Ergebnis: A cleaner, lower-confusion config.

Kombinationen

Mit anderen MCPs für 10-fache Wirkung

claude-code-organizer + jetski

Organizer audits your local config; Jetski audits server-side usage

After organizer trimmed memories, check Jetski's analytics for any drop in tool-call failures.✓ Kopiert
claude-code-organizer + marm-systems

Move long-lived memory from CC memories into a proper memory MCP

Find memories in CC that are just factual project notes; propose moving them into MARM-Systems via marm_contextual_log.✓ Kopiert

Werkzeuge

Was dieses MCP bereitstellt

WerkzeugEingabenWann aufrufenKosten
list_scopes (none) Full audit free
scan_mcps (none) or target server After adding any MCP, or periodically free (connects to each MCP briefly)
token_report scope? Before pruning free
dedupe (none) Config hygiene free

Kosten & Limits

Was der Betrieb kostet

API-Kontingent
None
Tokens pro Aufruf
Reports: 500-3000 tokens
Kosten in €
Free
Tipp
Run the scanner after every new MCP install — it's the cheapest defense.

Sicherheit

Rechte, Secrets, Reichweite

Credential-Speicherung: Reads your CC config files; doesn't transmit
Datenabfluss: Local dashboard only; connects to MCP servers only to read tool definitions

Fehlerbehebung

Häufige Fehler und Lösungen

Dashboard port in use

Kill the previous instance or set a different port via env.

Prüfen: lsof -i :<port>
Scanner can't reach a stdio MCP

The scanner spawns the same command your client does; make sure the command works from terminal.

Prüfen: Run the MCP command manually
False positive on an MCP you trust

Review the match, add it to an allowlist in organizer settings if it's clearly benign.

Prüfen: Re-run scan with allowlist

Alternativen

claude-code-organizer vs. andere

AlternativeWann stattdessenKompromiss
JetskiYou want server-side analytics and auth, not local config cleanupDifferent concern — runtime vs setup
Manual grep through ~/.claudeYou enjoy painNo scanner, no UI

Mehr

Ressourcen

📖 Offizielle README auf GitHub lesen

🐙 Offene Issues ansehen

🔍 Alle 400+ MCP-Server und Skills durchsuchen