xhs_content_agent

Content Agent — Xiaohongshu AI Content Assistant A Xiaohongshu content mining and automatic generation system built with FastAPI, LangChain, and OpenAI. Supports a complete closed loop from crawling competitor data, analyzing popular content patterns, AI-generated content and images, to publishing with one click on Xiaohongshu. ## Tool List | Module | Description | |------|------| | Data Crawling | Crawl Xiaohongshu search results using Playwright, supporting filtering by keywords, comments, likes, and collections, and automatically skipping videos and ads | | Data Analysis | Extract high-frequency keywords, popular tags,, and user insights output structured analysis reports Topic Generation | Based on analysis results, call LLM to generate several high-quality topic suggestions (including titles) | | Content Generation | Generate multiple complete texts for each topic: title, topic tags, interactive image suggestions, and content type Image Generation | Call OpenAI-image-1` to that fit Xiaohongshu and save them locally | | | Support MCP protocol (recommended REST API to publish content toongshu | | Feishu Sync | Synchronize crawled data-generated content to Feishu multi-dimensional tables for team collaboration and review | Server | Encapsulate the complete pipeline as an MCP tool, directly called in AI tools like Claude Desktop/Cursor | | Web-in static front-end, graphical operation interface for crawling, and publishing | ## Getting### 1. Install Dependenciesbash python -m venv .venv\Scripts\ Windows pip install -.txt playwright install chromium ``` ###. Configure Copy the `.env` file and fill in the necessary fields: ```env # OpenAI _API_KEY=sk-... _MODEL=gpt-4o-mini OPENAI_BASE_URL= # Optional, proxy address # Image Generation IMAGE_MODEL=gpt-image-1 # Fe-Dimensional Table (Optional) U_APP_ID= FEISHU_APP_SECRET= FEISHU= FEISHU= # Crawler data tableISHU_PUBLISH_TABLE # AI-generated note table # Xiaohong Service (Local) XCP_URL=http://localhost:18060 XHS_MCP_ENDPOINT=http://localhost:18060 ``` ### Start the Service ```bash uvicorn app.main:app --reload --host 127.0.0.1 --port 8000 ``` Access `http://127.0.0.1:8000` in your browser to enter the Web UI. FastAPI interactive documentation: `http://127.0.0.1:8000/docs` ## Main API Route | Description------|------| POST /analysis` | Analyze note return keywords, tags, title patterns, and insights | POST /topics/generate Generate topic suggestions based on | | `content/generate Generate text and based on| `POST /agent/run` | Run the complete content generation pipeline (analysis → topic → content) with | | `POST /crawl/search` | Crawl Xiaohongshu text notes by keyword | | /publish/prepare` | Assemble publish payload (REST/M) | | `POST /publish/send Publish to Xiaohongshu | | `POST //sync` | Synchronize generated content to Feishu | | `POST /feishu/sync-craw | Synchronize crawled data to Fe| /health` | Health check | ## Using MCP Server Encapsulate this project as an MCP Server, which can be registered and used in AI tools that support the MCP protocol (Claude Desktop, Cursor, etc.). python mcp_server.py ``` Provide the following MCP| Tool | Description | |------|------| | `run_content_pipeline` | Run the generation pipeline | | `generate_xhs_images` | based on| `_xhs` | Generate images and publish to Xiaohongshu with one click | | `check_xhs_login` Xiaohongshu | ## Project Structure ``` xhs_content_agent/ ├── app/ │ ├── api/ # FastAPI routing layer │ ├── core # Configuration (Settings) │ models/ # P data models ├── prompts/ LLM Prompt templates └── services/ # Business logic layer │ agent_service.py # Main pipeline orchest│ ├── analysis_service.py data analysis │ ├── topic_service.py # Topic generation │ content_service.py # Content generation │ ├── image_service.py Image generation │ publish_service.py Xiaohong │ ├── feishu_service.py # Feishu synchronization ├── localrawler_service.py # Xiaohongshu crawler │ └── mcp_client_service.py # MCP client ├── static/ # Web front-end pages ├── data/ │ ├── raw/ # Crawled data/sample data │ └── output/images # Generated images ├── mcp_server.py Server entry .txt .env # Environment variable configuration Technology Stack Backend Framework**: FastAPI + Uvicorn **LLM Call**: LangChain +-OpenAI (GPT-4o-mini **Image Generation**: OpenAI gpt-image-1 - **Crawler**: Playwrightromium) - **MCP Protocol**: `mcp` SDK (FastMCP) - **Feishu API**: Feishu Multi-Dimensional Table Open API - **Chinese Processing**: word segmentation - **Data Validation**: Pydantic v2 ## Precautions - The crawling function needs to complete Xiaoh login in advance and save Cookies to `data/raw/xhs_cookies.json`. - The publishing function relies on the local Xiaohongshu MCP service (default port `18060`), which requires completing the scan code login. - Image generation requires an OpenAI API Key with `gpt-image-1` access permissions. - Feishu synchronization is an optional If not configured, related interfaces will return prompt information without errors.