ue5-docs-mcp

# UE5 Docs — Offline Documentation & LLM Context for Unreal Engine 5.7 [](https://github.com/chcardenasto/DocUnrealEngine/actions/workflows/deploy.yml) [](https://dev.epicgames.com/documentation/en-us/unreal-engine) [](docs-md-py/en-us/unreal-engine) [](LICENSE) [](https://python.org) > **3,444 pages of Unreal Engine 5.7 official documentation in clean Markdown — purpose-built for LLMs, RAG pipelines, offline reading, and AI-assisted development.** --- ## Why this project? Official Unreal Engine documentation lives behind a JavaScript-heavy web interface. This project downloads the entire documentation corpus and converts it to clean, structured Markdown — the native format preferred by LLMs and embedding models. | Format | Use case | |--------|----------| | 3,444 individual `.md` files | RAG pipelines, AI code assistants, semantic search | | `GUIDE_BOOK.md` (~30 MB) | Full-corpus context, local LLM input, `Ctrl+F` search | | `GUIDE_INDEX.md` | Hierarchical overview, LLM navigation hints | | MkDocs site | Interactive offline browser with full-text search | | **MCP Server** | Direct integration with Claude Code and any MCP-compatible LLM | --- ## Quick Start ### Option 1 — MCP Server (for Claude Code users) Install the MCP server so Claude can search and read UE5 docs in any session: ```bash # 1. Clone the repository git clone https://github.com/chcardenasto/DocUnrealEngine.git cd DocUnrealEngine # 2. Install the MCP server package pip install -e ".[mcp]" # 3. Register it with Claude Code claude mcp add --transport stdio ue5-docs -- python -m mcp_server.server --docs-dir ./docs-md-py ``` Then in any Claude Code session: ``` What is the Gameplay Ability System in Unreal Engine? How do I implement replication for a custom Actor? ``` Claude will automatically query the MCP server for relevant documentation. ### Option 2 — Interactive Web Browser (MkDocs) ```bash pip install -r requirements.txt python scripts/build_guide.py --mode mkdocs --docs-dir docs-md-py mkdocs serve # Open http://127.0.0.1:8000 ``` ### Option 3 — RAG Pipeline Input The `docs-md-py/en-us/unreal-engine/` directory contains 3,444 clean Markdown files ready to embed: ```python from langchain_community.document_loaders import DirectoryLoader from langchain_text_splitters import MarkdownHeaderTextSplitter loader = DirectoryLoader("./docs-md-py", glob="**/*.md", show_progress=True) docs = loader.load() splitter = MarkdownHeaderTextSplitter(headers_to_split_on=[("#", "H1"), ("##", "H2")]) chunks = [chunk for doc in docs for chunk in splitter.split_text(doc.page_content)] # Feed chunks into ChromaDB, Pinecone, FAISS, etc. ``` ### Option 4 — Offline Index (No server) ```bash python scripts/build_guide.py --mode index --docs-dir docs-md-py # Opens GUIDE_INDEX.md in any Markdown viewer (VS Code, Obsidian, etc.) ``` --- ## MCP Server Details The included MCP server exposes three tools to any MCP-compatible LLM: | Tool | Description | |------|-------------| | `search_docs` | Search docs by keyword, topic, or class name | | `get_doc` | Read the full content of a specific page | | `list_sections` | Browse top-level documentation categories | **Installation options:** ```bash # stdio transport (local) claude mcp add --transport stdio ue5-docs -- python -m mcp_server.server # With custom docs path claude mcp add --transport stdio ue5-docs -- python -m mcp_server.server --docs-dir /path/to/docs-md-py # Via pip (once published) pip install ue5-docs-mcp claude mcp add --transport stdio ue5-docs -- python -m ue5_docs_mcp ``` --- ## Dual MCP Setup — Docs + Unreal Editor (UE 5.8+) > Unreal Engine 5.8 ships with a built-in MCP server inside the editor. Combine it with this docs server for a complete AI-assisted workflow. **With both servers active, Claude can:** 1. **Look up documentation** (this server) — *"What parameters does SpawnActorDeferred take?"* 2. **Control the Unreal Editor** (Epic's server) — *"Spawn a PointLight at (100, 0, 500) and set intensity to 5000"* **Setup:** 1. Enable the `Unreal MCP` plugin in UE 5.8+ 2. Run `ModelContextProtocol.GenerateClientConfig ClaudeCode` in the Output Log 3. Add our docs server to the generated `.mcp.json` — see [`.mcp.json.template`](.mcp.json.template) Full setup guide in [`LLM_INTEGRATION.md`](LLM_INTEGRATION.md). --- ## LLM Integration Guide See [`LLM_INTEGRATION.md`](LLM_INTEGRATION.md) for detailed strategies: - **Cursor / GitHub Copilot / Cline** — Index the `docs-md-py` folder as a workspace - **Google NotebookLM / Custom GPTs / Claude Projects** — Upload the ZIP of `docs-md-py` - **RAG pipelines (LangChain, LlamaIndex)** — Embed individual `.md` files - **Large context models (Gemini 1.5 Pro, Claude)** — Feed `GUIDE_INDEX.md` first, then specific pages --- ## Versioning Each Unreal Engine release is frozen as a **git tag**. `main` always contains the latest version. | Tag | UE Version | Status | |-----|-----------|--------| | `v5.7` | Unreal Engine 5.7 | Current (3,444 pages) | | `v5.8` | Unreal Engine 5.8 | Planned | ```bash # Use UE 5.7 docs (frozen) git checkout v5.7 # Use latest docs git checkout main ``` --- ## Updating or Scraping a New Version ```bash # Scrape UE 5.8 (when released) python scripts/scrape_epic_docs.py --out-dir docs-md-py --workers 4 --application-version 5.8 # Resume an interrupted download python scripts/scrape_epic_docs.py --out-dir docs-md-py --workers 4 --skip-existing # Force re-download all pages python scripts/scrape_epic_docs.py --out-dir docs-md-py --force ``` **Scraper features:** - Multi-threaded (configurable workers) - Resumable — skips already-downloaded pages - Exponential backoff on failures - Generates `INDEX.md`, `FAILURES.md`, and `pending-routes.json` on completion --- ## Repository Structure ``` DocUnrealEngine/ ├── docs-md-py/ │ └── en-us/unreal-engine/ ← 3,444 Markdown files (the corpus) │ ├── INDEX.md ← Hierarchical page index │ ├── FAILURES.md ← 26 pages that returned 404 │ └── mkdocs.yml ← Auto-generated MkDocs config ├── mcp_server/ │ ├── __init__.py │ └── server.py ← MCP server (search + retrieval) ├── scripts/ │ ├── scrape_epic_docs.py ← Web scraper (HTML → Markdown) │ └── build_guide.py ← Compiler (index / book / mkdocs) ├── .github/workflows/deploy.yml ← CI/CD → GitHub Pages ├── GUIDE_BOOK.md ← Full corpus as single file (~30 MB) ├── GUIDE_INDEX.md ← Hierarchical index ├── LLM_INTEGRATION.md ← Detailed LLM usage guide ├── llms.txt ← LLM discovery metadata ├── pyproject.toml ← Package config for MCP server ├── requirements.txt ← Python dependencies └── README.md ``` --- ## Documentation Coverage | Metric | Value | |--------|-------| | UE version | 5.7 | | Pages downloaded | 3,444 | | Failed pages (404) | 26 (deprecated tools/removed sections) | | Success rate | 99.25% | | Corpus size | ~40 MB (individual files) | | GUIDE_BOOK.md | ~30 MB | The 26 failed pages are all confirmed deprecated or removed sections from Epic's servers (e.g., `vertex-sculpt-tool`, `dynamic-sculpt-tool`, `variant-manager-template-overview`). --- ## Contributing See [`CONTRIBUTING.md`](CONTRIBUTING.md) for guidelines. In short: - Open an issue before implementing a large change - Keep PRs focused and small - Run `pip install -r requirements.txt` before working on scripts --- ## Legal Notice The documentation content in `docs-md-py/` is the intellectual property of **Epic Games, Inc.** and is subject to Epic Games' terms of use and copyright. This repository provides open-source tooling (scraper, builder, MCP server) for offline access and personal/educational use only. It does not distribute commercial content or claim authorship of the original documentation. --- ## License The **scripts and tooling** in this repository (everything except the content in `docs-md-py/`) are licensed under the [MIT License](LICENSE). The **documentation content** in `docs-md-py/` remains the property of Epic Games, Inc.