Giving AI agents hands without handing over the keys: how we think about safety at TwinMCP

Description and use

<p><strong>TwinMCP</strong>, a hosting platform for MCP servers. This post explains the safety choices we made, including the ones we haven't solved yet. Criticism welcome.</p> <p>With MCP, ChatGPT, Claude or Cursor stop being chatbots. They read your Slack, write to your Notion, query your database, list your Stripe invoices. The risk moves from what the model says to what the model does.</p> <p>So when we built a platform that hosts MCP servers for other people, "AI safety" stopped being an abstract debate for us. It became a list of very concrete questions: what can this agent touch, with w

Problem

<p><strong>TwinMCP</strong>, a hosting platform for MCP servers. This post explains the safety choices we made, including the ones we haven't solved yet. Criticism welcome.</p> <p>With MCP, ChatGPT, Claude or Cursor stop being chatbots. They read your Slack, write to your Notion, query your database, list your Stripe invoices. The risk moves from what the model says to what the model does.</p> <p>So when we built a platform that hosts MCP servers for other people, "AI safety" stopped being an abstract debate for us. It became a list of very concrete questions: what can this agent touch, with w

Outcome

<p><strong>TwinMCP</strong>, a hosting platform for MCP servers. This post explains the safety choices we made, including the ones we haven't solved yet. Criticism welcome.</p> <p>With MCP, ChatGPT, Claude or Cursor stop being chatbots. They read your Slack, write to your Notion, query your database, list your Stripe invoices. The risk moves from what the model says to what the model does.</p> <p>So when we built a platform that hosts MCP servers for other people, "AI safety" stopped being an abstract debate for us. It became a list of very concrete questions: what can this agent touch, with w

Evidence

Giving AI agents hands without handing over the keys: how we think about safety at TwinMCP

Source