# Claude Desktop Integration Fix
## Current Situation
- Production Qdrant is running on port 6333
- Test SearXNG is running on port 8081
- The MCP server can use these existing services
## Solution
The test environment (`.env.test`) is already configured correctly:
- `TRANSPORT=stdio` (for Claude Desktop)
- `QDRANT_URL=http://localhost:6333` (using production Qdrant)
- `SEARXNG_URL=http://localhost:8081` (using test SearXNG)
## Testing the Integration
1. **Verify services are running:**
```bash
docker ps | grep -E "(qdrant|searxng)"
```
2. **Test the MCP server locally:**
```bash
export USE_TEST_ENV=true
uv run python src/crawl4ai_mcp.py
```
You should see:
- "Transport mode: stdio"
- "Running with STDIO transport"
3. **Send a test initialization:**
```json
{"jsonrpc": "2.0", "method": "initialize", "params": {"protocolVersion": "0.1.0", "capabilities": {}, "clientInfo": {"name": "test", "version": "1.0"}}, "id": 1}
```
## Claude Desktop Configuration
Use this configuration in `%APPDATA%\Claude\claude_desktop_config.json`:
```json
{
"mcpServers": {
"crawl4ai-rag": {
"command": "wsl",
"args": [
"--cd",
"/home/krashnicov/crawl4aimcp",
"--",
"bash",
"-c",
"USE_TEST_ENV=true /home/krashnicov/.local/bin/uv run python src/crawl4ai_mcp.py"
]
}
}
}
```
## Troubleshooting
If the MCP server isn't working in Claude Desktop:
1. **Check the OpenAI API key in `.env.test`:**
- Currently set to `test-mock-key-not-real`
- You may need to add a real API key for embedding generation
2. **Check service connectivity:**
```bash
curl http://localhost:6333/healthz # Qdrant
curl http://localhost:8081/healthz # SearXNG
```
3. **Review logs:**
- Check Claude Desktop developer console for errors
- Run the MCP server manually to see detailed output
## Summary
You don't need to run `docker-compose.test.yml` if you're just using the MCP server with Claude Desktop. The existing services (production Qdrant on 6333 and test SearXNG on 8081) are sufficient.
