LangLa Server MCP
# LangLa Server MCP — cho agent sửa client
MCP stdio cung cấp **docs protocol server** (`CLIENT_*.md`, `SERVER_*.md`) để agent LibGDX (`LangLa_kl`) tra wire MSG mà không đoán packet.
## Tools
| Tool | Việc |
|------|------|
| `list_docs` | Liệt kê tài liệu protocol |
| `read_doc` | Đọc 1 file docs (vd. `CLIENT_STATS_LONG.md`) |
| `search_docs` | Tìm chuỗi / MSG trong toàn bộ docs |
| `get_msg` | Tra nhanh theo số MSG (vd. `62`, `-49`, `-88`) |
| `server_overview` | Tóm tắt cổng, long money/stats, thứ tự ưu tiên đọc docs |
## Cài
```bash
cd langla-mcp
npm install
```
## Cursor — `mcp.json`
```json
{
"mcpServers": {
"langla-server": {
"command": "node",
"args": ["/ABS/PATH/langla-mcp/src/index.js"],
"env": {
"LANGLA_DOCS_DIR": "/ABS/PATH/langla-mcp/docs"
}
}
}
}
```
Hoặc trỏ `LANGLA_DOCS_DIR` thẳng vào `langla-server/docs` để luôn bản mới nhất.
## Sync docs từ server
```bash
# từ máy có clone langla-server
cp ../langla-server/docs/CLIENT_*.md docs/
cp ../langla-server/docs/SERVER_*.md docs/
cp ../langla-server/docs/AUDIT_*.md docs/
```
## Không làm
- Không đổi cổng 2907/2908, MySQL, thứ tự field packet.
- Không đoán `read*`/`write*` — luôn `read_doc` / `get_msg` trước.
TDQS
Scored across 5 tools
Each tool targets a distinct access pattern: listing files, reading a whole file, searching free text, and looking up a specific MSG number. read_doc and get_msg overlap slightly since both return document content, but get_msg is clearly scoped to MSG-numbered entries, so an agent can choose reliably.
Four of five tools follow a clean verb_noun pattern (read_doc, list_docs, search_docs, get_msg). server_overview breaks the pattern by being noun-only, a minor but noticeable deviation.
Five tools is well-scoped for a documentation-retrieval server, with each tool covering a distinct lookup strategy. No redundant or filler tools.
The surface covers the core docs workflow: overview, listing, full read, free-text search, and targeted MSG lookup. Minor gaps exist, such as no way to enumerate all MSG entries or page through very large documents, but agents can work around these via search_docs.