{
  "markdown": "# Open-Crab 🦀\n\nSelf-maintaining system toolkit for Claude Code. Set up the ecosystem and it runs itself.\n\n## Commands\n\n| Command | Description | Daily |\n|---------|-------------|-------|\n| `/daily` | Run all `daily: true` commands, output summary table, log results | - |\n| `/disk-clean` | Scan current directory for large files and cleanable caches | No |\n| `/security` | System security scan (ports, users, services, sensitive files) | Yes |\n| `/evolve` | Self-review all commands, check scan history, fix issues | No |\n\n## MCP Tools\n\nOffload file understanding and content analysis to a local LLM, saving Claude API tokens.\n\n| Tool | Description |\n|------|-------------|\n| `file_summarize` | Read a file and generate a summary |\n| `content_qa` | Ask questions about text content |\n| `file_search` | Search files and rank by relevance |\n| `scan_log_analyze` | Analyze scan_log trends and patterns |\n| `llm_status` | Check local LLM server status |\n\nSupports any OpenAI-compatible backend (Ollama, LM Studio, llama.cpp, vLLM, etc.) via environment variables.\n\n## How it works\n\n```\n/daily ──→ scans commands with `daily: true` ──→ runs them ──→ logs to scan_log.md\n              │\n/evolve ──→ reviews all commands ──→ reads scan_log.md\n              │                         │\n              ├─ fixes outdated commands │\n              ├─ detects OS compatibility │\n              └─ checks recurring errors  │\n                                          ▼\n                               self-healing loop\n\nClaude Code ←─stdio─→ MCP Server ←─HTTP─→ Local LLM (e.g. Ollama)\n                        │\n                        ├─ file understanding\n                        ├─ content analysis\n                        └─ relevance ranking\n```\n\n## Install\n\n### Commands only (no MCP)\n\n```\n/install Open-Crab\n```\n\nOr clone manually:\n\n```bash\ngit clone https://github.com/jay2zxy/Open-Crab.git ~/.claude/plugins/Open-Crab\n```\n\nCommands work immediately — no extra setup needed.\n\n### Commands + MCP Tools\n\n1. Install a local LLM backend (pick one):\n   - [Ollama](https://ollama.com) (recommended) — `ollama pull qwen3:8b`\n   - [LM Studio](https://lmstudio.ai) — download a model, start server\n   - Any OpenAI-compatible server (llama.cpp, vLLM, etc.)\n\n2. Install MCP dependencies:\n\n```bash\ncd ~/.claude/plugins/Open-Crab/mcp && npm install\n```\n\n3. Test the connection:\n\n```bash\nnode mcp/llm.js\n```\n\n4. If not using Ollama defaults, edit `.mcp.json` in the plugin root:\n\n```json\n{\n  \"env\": {\n    \"LLM_BASE_URL\": \"http://localhost:1234/v1\",\n    \"LLM_MODEL\": \"your-model-name\"\n  }\n}\n```\n\n## Requirements\n\n- Claude Code CLI\n- Works on Windows / macOS / Linux\n- (Optional) Node.js + local LLM server for MCP tools\n\n### MCP Configuration\n\n| Variable | Default | Description |\n|----------|---------|-------------|\n| `LLM_BASE_URL` | `http://localhost:11434/v1` | OpenAI-compatible API endpoint |\n| `LLM_MODEL` | `qwen3:8b` | Model to use |\n| `LLM_API_KEY` | `no-key` | API key (if required) |\n\n## Customization\n\n- Add `daily: true` to any command's frontmatter to include it in daily patrol\n- Add new commands to `commands/` and `/evolve` will automatically review them\n- Scan logs are stored in the plugin directory (`scan_log.md`, gitignored)\n",
  "bytes": 3226,
  "sha": "4cc9167166c9115b4b4ecd75dbc3e03e766d491b8addc90d54e1f4dc09abbc38",
  "repo_slug": "jay2zxy/open-crab",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_jay2zxy_open_crab_open_crab_d9cec7cf/readme"
}