# Slander MCP

An MCP server that finds humorous roasts, jokes, and memes about any character (real or fictional) by searching Twitter/X. Uses social proof (engagement metrics) to rank content, with targeted LLM involvement for query generation, batch quality assessment, and nickname extraction.

## Installation

```bash
npm install
npm run build
```

## Configuration

Create a `.env` file with:

```bash
# Twitter/X API credentials (required)
TWITTER_BEARER_TOKEN=your_twitter_bearer_token_here

# LLM API key (at least one required)
ANTHROPIC_API_KEY=your_anthropic_api_key_here
# or
OPENAI_API_KEY=your_openai_api_key_here
```

## Usage with Claude Desktop

Add to your Claude Desktop MCP configuration:

```json
{
  "mcpServers": {
    "slander": {
      "command": "node",
      "args": ["/path/to/slander_mcp/dist/index.js"],
      "env": {
        "TWITTER_BEARER_TOKEN": "your_token",
        "ANTHROPIC_API_KEY": "your_key"
      }
    }
  }
}
```

## Tools

### `generate_search_query`

Generate effective Twitter search queries for finding slander about a target.

**Input:**
- `target` (string, required): Name of character to search for

**Output:**
- `queries`: Array of search query strings

**Example:**
```json
Input:  { "target": "LeBron James" }
Output: { "queries": ["LeBron James ratio", "LeChoke", "LeBron hairline", ...] }
```

### `fetch_posts`

Fetch posts from Twitter for a given query, looping until quality threshold is met.

**Input:**
- `query` (string, required): Search query
- `loop_limit` (number, optional): Max fetch iterations (default: 5)
- `count` (number, optional): Posts per fetch (default: 10)
- `target` (string, optional): Target name for quality evaluation

**Output:**
- `posts`: Array of post objects with engagement metrics
- `iterations`: Number of fetch loops
- `stopped_reason`: "quality_threshold" or "loop_limit"

### `rank_posts`

Rank fetched posts by engagement, separate text from media, extract nicknames.

**Input:**
- `posts` (array, required): Posts from fetch_posts
- `top_n` (number, optional): Results per category (default: 3)
- `target` (string, optional): Target name for nickname extraction

**Output:**
- `text_posts`: Top text posts ranked by engagement
- `media_posts`: Top media posts ranked by engagement
- `nicknames`: Extracted nicknames/slang for the target

**Engagement Score Formula:**
```
score = (likes * 1.0) + (retweets * 2.0) + (replies * 0.5)
```

## Example Workflow

1. Generate search queries:
   ```
   generate_search_query({ target: "LeBron James" })
   ```

2. Fetch posts for each query:
   ```
   fetch_posts({ query: "LeChoke", target: "LeBron James" })
   ```

3. Combine and rank results:
   ```
   rank_posts({ posts: [...all_posts], top_n: 5, target: "LeBron James" })
   ```

## License

MIT