Most [MCP servers](https://github.com/modelcontextprotocol/servers) suggest using `npx -y` as the recommended way to install a server. This downloads and executes arbitrary scripts from the internet. This is grossly insecure and I think the MCP authors sh
Most [MCP servers](https://github.com/modelcontextprotocol/servers) suggest using `npx -y` as the recommended way to install a server. This downloads and executes arbitrary scripts from the internet. This is grossly insecure and I think the MCP authors sh
npx -y @considered/harmfulAdd this configuration to your claude_desktop_config.json:
{
"mcpServers": {
"considered-harmful-npm": {
"command": "npx",
"args": [
"-y",
"npx -y @considered/harmful"
]
}
}
}Restart Claude Desktop, then ask:
"What tools do you have available from harmful?"
API Key Required
This server requires an API key from harmful. Add it to your environment or config.
| Variable | Required | Description |
|---|---|---|
| HARMFUL_API_KEY | Yes | Your harmful API key |
"What resources are available in harmful?"
Claude will query available resources and return a list of what you can access.
"Show me details about [specific item] in harmful"
Claude will fetch and display detailed information about the requested item.
"Create a new [item] in harmful with [details]"
Claude will use the appropriate tool to create the resource and confirm success.
See what tools in your stack can connect to AI.
We build custom MCP integrations for B2B companies. From simple connections to complex multi-tool setups.