Skip to main content
Glama
Duela MCP Server

Duela MCP Server

Official
by duela-labs
README.md
# duela-mcp

> **Razvoj (Docker):** iz ovog foldera `docker compose up --build --watch mcp` — digne mcp i sve od čega zavisi (baza, Redis, api…). `compose.yaml` ovde samo uključuje `../duela-infra/compose.dev.yaml`, gde su opisani svi servisi. Komande ispod su za rad van Dockera.

MCP server for [Duela](https://duela.app): give your own AI agent (Claude Code, Codex, OpenCode or any MCP client) market data, technical indicators computed by Duela, news, and a place to save reports in your Duela dashboard.

Your AI runs on your own account. Duela only serves data and stores the reports you save.

## Tools

| Tool | What it does |
| --- | --- |
| `search_instruments` | find a stock, ETF, crypto asset or Serbian investment fund |
| `get_prices` | daily OHLCV prices |
| `get_indicators` | RSI, MACD, SMA, EMA, Bollinger, ATR — computed by Duela, not estimated by the model |
| `get_news` | summarized news with sources |
| `save_report` | save your analysis to your Duela dashboard |

## Connect

Create an MCP token in the Duela dashboard (Settings → MCP).

**Claude Code (hosted server):**

```bash
claude mcp add --transport http duela https://mcp.duela.app/mcp --header "Authorization: Bearer <TOKEN>"
```

**Local (stdio):**

```bash
DUELA_TOKEN=<TOKEN> npx @duela/mcp
```

Example prompt: *"Use duela to run a bull/bear analysis of NVDA with RSI and MACD and last week's news, then save it."*

## Rules built into the server

The server tells your agent: general analysis only, no buy/sell recommendations or quantities, and cite only numbers returned by the tools.

## Development

```bash
pnpm install
DUELA_TOKEN=dev DUELA_API_URL=http://localhost:4000 pnpm dev   # stdio
MCP_HTTP_PORT=8080 DUELA_API_URL=http://localhost:4000 pnpm start   # HTTP after pnpm build
```

This repository is public and does not depend on private Duela packages.

## License

MIT