Blog
How AI assistants work with real accounts: MCP, security and practical requests.
Remote vs local MCP servers: which should you use?
Local MCP servers run on your computer over stdio; remote ones run in the cloud over Streamable HTTP with OAuth. When each makes sense, and why most assistants are moving to remote.
November 4, 2026Prompt injection: how to let an AI read your email safely
An email can contain instructions aimed at your AI assistant. What prompt injection is, why it matters once an AI can send mail, and the four safeguards to look for in an MCP server.
November 4, 2026Too many MCP tools make your AI worse. Here is how to fix it
Every MCP tool’s name, description and input schema is sent to the model before you type. With a dozen servers that adds up. Why it hurts tool choice, and three ways to keep the list small.
November 4, 2026When should an AI assistant ask before it acts?
Reading is safe to automate; sending, deleting and paying are not. A simple way to sort AI actions into read, change and risky, and how approvals work in practice.
November 4, 202615 things to ask your AI once it can see your email and calendar
Practical requests for Claude, ChatGPT or Cursor once your mail, calendar and tasks are connected: triage, follow-ups, scheduling, travel, invoices and weekly reviews.
November 4, 2026OAuth for MCP: why your AI should never see your password
How remote MCP servers sign you in with OAuth 2.1, what the AI app actually receives, and how to revoke access. A plain-language guide for anyone connecting accounts to Claude or ChatGPT.
Try it with your own accounts
Free plan, no card. Setup takes a couple of minutes.