MDCP — Master Developer Claude Plugin
A modular developer toolkit for coding agents. MDCP combines curated skills, platform-specific instruction adapters, optional MCP connector guidance, and a local project-memory utility designed to reduce repeated context preparation.
V2000 architecture track: a roadmap/version label, not a claim that 2,000 files, agents, plugins, or integrations are implemented. Features are counted only when implemented and validated.
Explore the skills visually
The official MDCP website source is maintained in a separate repository. Its Cloudflare Pages build generates the searchable, phone-friendly skill catalog directly from this repository's skills/*/SKILL.md files.
YouTube Creator Ecosystem
MDCP includes a cloud-ready YouTube Studio toolkit for channel strategy, video editing, Shorts, thumbnails, SEO, approval-first publishing, analytics and monitoring. See the YouTube ecosystem guide, plugin runtime, and upstream repository catalogue.
Unified Creator + Cybersecurity Automation
MDCP adds one compact cross-agent umbrella skill for authorized cybersecurity defense, website vulnerability review, SQL injection prevention, E2E testing, and complete rights-safe YouTube automation. The unified router is documented in skills/creator-security-automation/SKILL.md, with portable mirrors under .agents/skills/ and .gpt/skills/. Its single MCP tool adapter is in plugins/mdcp-unified/: passive HTTPS/header audit, heuristic SQL source review, E2E test-plan generation, and delegation to the existing YouTube runtime when configured.
Security checks are defensive and authorization-bound; they do not perform exploit attempts or bypass access controls. YouTube publishing and other external mutations remain approval-gated. Thumbnail generation, editing/rendering, cloud queues, OAuth and remote hosting require their respective configured tools/services; documentation does not imply those services are already deployed.
Existing skill library
MDCP currently includes 91 curated skills across core development, security, backend/data, frontend, design, Three.js, workflows, clean-room app reconstruction, and skill/plugin/MCP authoring. See the skill catalog and third-party notices for source and license details.
Platform coverage
| Platform | Included guidance | Important boundary |
|---|
| Claude Code | Existing marketplace plugin and Claude adapter | Native plugin/skill installation |
| OpenAI Codex | AGENTS.md and Codex adapter | Host must support the documented instruction discovery |
| ChatGPT | Project / Custom GPT setup guide | Not an automatically installed ChatGPT plugin |
| Gemini | Portable GEMINI.md adapter | Add to a supported workspace manually |
| Cursor | .cursorrules starter | Follow current Cursor rules support |
| Windsurf | .windsurfrules starter | Follow current Windsurf rules support |
| Other agents | Portable Markdown context | Adapt to the host's own format |
See platform adapters and the universal architecture.
Using Claude on Android or iPhone? Follow the Claude.ai mobile-first setup guide for Project instructions, skill uploads, remote MCP connectors, and phone-friendly workflows.
Token efficiency and project memory
LLM inference is not free and MDCP cannot guarantee zero token consumption or a fixed 80–90% reduction. It helps avoid unnecessary repeated reads by creating a small, bounded project-context pack and storing explicit decisions locally.
Quick start
# Build a compact context pack (stdout)
python3 tools/mdcp_memory.py pack --root . --max-chars 12000
# Save a project decision locally
python3 tools/mdcp_memory.py remember "API versioning" "Use /v1 routes for public endpoints."
# Search saved decisions
python3 tools/mdcp_memory.py search "API"
# Export or delete a memory
python3 tools/mdcp_memory.py export
python3 tools/mdcp_memory.py forget 1
The database lives in .mdcp/memory.sqlite3 and is ignored by Git by default. The utility does not write to ChatGPT Memory, Claude memory, Codex session memory, or any provider-private storage. Feed the generated context pack to the agent or configure the host to load it. Read Token Efficiency & Memory.
Free and local LLM routing (OmniRoute + Ollama)
MDCP includes an integration pack for OmniRoute, a separately maintained open-source AI gateway. It provides one local OpenAI-compatible endpoint, provider/model routing, and fallback according to the providers and accounts you configure. MDCP does not vendor or claim ownership of OmniRoute.
여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗
