# mcp-server-deerflow-kinthai

MCP Server that exposes [DeerFlow](https://github.com/bytedance/deer-flow) deep capabilities via standard [Model Context Protocol](https://modelcontextprotocol.io/).

Any MCP client (OpenClaw, Claude Desktop, Cursor, etc.) can discover and invoke DeerFlow skills through this server.

## Architecture

```
MCP Client (OpenClaw / Claude Desktop / Cursor)
    |
    | MCP protocol (SSE on :8808)
    v
mcp-server-deerflow-kinthai
    |
    | LangGraph REST API (:2024)
    v
DeerFlow (bytedance/deer-flow)
    |
    +-- deep research (multi-source web search + cross-verification)
    +-- data analysis (DuckDB)
    +-- chart visualization (26+ chart types)
    +-- PPT generation
    +-- image generation
    +-- consulting analysis (SWOT, Porter's, etc.)
```

The server is a thin wrapper: it translates MCP tool calls into DeerFlow LangGraph runs, extracts the response text and artifacts, and returns them in MCP format. DeerFlow itself remains untouched upstream.

## Tools

| Tool | Description |
|------|-------------|
| `deep_research` | Multi-source web research with cross-verification |
| `data_analysis` | Data analysis with DuckDB (CSV/Excel) |
| `chart_visualization` | 26+ chart types (line, bar, pie, scatter, sankey, etc.) |
| `ppt_generation` | PowerPoint presentation generation |
| `image_generation` | AI image generation |
| `consulting_analysis` | Business analysis (SWOT, Porter's Five Forces, etc.) |

All tools accept a `query` string (required) and an optional `agent_name` for specialized DeerFlow agent personas.

## Quick Start

```bash
# Install
pip install mcp-server-deerflow-kinthai

# Run (requires a running DeerFlow instance)
export DEERFLOW_LANGGRAPH_URL=http://localhost:2024
mcp-server-deerflow-kinthai
```

The server starts on port 8808 with SSE transport at `/sse`.

Requires Python >= 3.12.

## Prerequisites

You need a running [DeerFlow](https://github.com/bytedance/deer-flow) instance. Follow the DeerFlow README to set it up, then point this server at it:

```bash
# Default: DeerFlow LangGraph on localhost:2024
export DEERFLOW_LANGGRAPH_URL=http://localhost:2024

# Optional: DeerFlow Gateway for artifact downloads (charts, PPTs, images)
export DEERFLOW_GATEWAY_URL=http://localhost:8001
```

## Configuration

### Environment Variables

| Variable | Default | Description |
|----------|---------|-------------|
| `DEERFLOW_LANGGRAPH_URL` | `http://localhost:2024` | DeerFlow LangGraph server URL |
| `DEERFLOW_GATEWAY_URL` | `http://localhost:8001` | DeerFlow Gateway API URL (for artifact downloads) |

### OpenClaw

Add to your `openclaw.json`:

```json
{
  "mcp": {
    "servers": {
      "deerflow-kinthai": {
        "url": "http://localhost:8808/sse"
      }
    }
  }
}
```

### Claude Desktop

Add to your Claude Desktop config:

```json
{
  "mcpServers": {
    "deerflow-kinthai": {
      "command": "mcp-server-deerflow-kinthai"
    }
  }
}
```

### Claude Code

```bash
claude mcp add deerflow-kinthai http://localhost:8808/sse --transport sse
```

## Mount in Existing App

The server can be embedded in an existing FastAPI/Starlette application:

```python
from fastapi import FastAPI
from mcp_server_deerflow_kinthai.server import create_starlette_app

app = FastAPI()
app.mount("/mcp", create_starlette_app())
```

## Development

```bash
git clone https://github.com/kinthaiofficial/mcp-server-deerflow-kinthai
cd mcp-server-deerflow-kinthai
pip install -e ".[dev]"
pytest
```

## Related Projects

- [DeerFlow](https://github.com/bytedance/deer-flow) — The upstream multi-agent research framework by ByteDance
- [openclaw-kinthai](https://github.com/kinthaiofficial/openclaw-kinthai) — OpenClaw channel plugin for KinthAI
- [kinthai-agent-cli](https://github.com/kinthaiofficial/kinthai-agent-cli) — Universal CLI bridge for connecting any agent to KinthAI

## License

MIT — [KinthAI](https://kinthai.ai)