May 17, 2026 · 8 min read
How to add Helix to ChatGPT with MCP connectors
Connect Helix to ChatGPT as a custom MCP connector so ChatGPT can summarize email, draft replies, inspect calendar context, and leave sensitive actions pending approval.
Blog
Engineering and design notes.
Helix engineering and design notes. What we ship, what we cut, and what we have learned about giving AI agents a real inbox with human approval on every send.
May 17, 2026 · 7 min read
How to add Helix to Claude Code
Set up Helix in Claude Code with the Helix MCP server so your terminal coding agent can use email, calendar, contacts, and approval-aware workflow context.
May 17, 2026 · 7 min read
How to add Helix to Cursor as a remote MCP server
Connect Helix to Cursor through a remote MCP server so your AI code editor can use live email, calendar, contacts, and identity context while keeping actions approval-gated.
May 16, 2026 · 8 min read
Best AI email assistant for teams: what to look for before you deploy one
Choosing the best AI email assistant for a team is less about clever drafting and more about permissions, approvals, audit logs, revocation, and safe handoff to real workflows.
May 15, 2026 · 9 min read
How to build an AI email agent with MCP without giving it your whole inbox
An MCP email agent should not need unlimited mailbox access. Use scoped inbox tools, approval gates, audit logs, and revocation so agents can act safely in production.
May 14, 2026 · 7 min read
Human-in-the-loop AI email approval: the pattern every agent needs
Human-in-the-loop AI email approval is not a speed bump. It is the control surface that lets agents draft, schedule, and follow up while humans keep authority.
May 13, 2026 · 8 min read
Gmail and Outlook AI agent security: safer inbox automation patterns
Gmail and Outlook AI agents need more than OAuth consent. Safer inbox automation uses scoped access, dedicated agent addresses, approval gates, and revocation.
May 12, 2026 · 7 min read
AI agent audit logs for email and calendar workflows
AI agent audit logs should capture source context, reasoning, approvals, policy matches, email sends, calendar writes, and revocation events in one reviewable trail.
May 1, 2026 · 8 min read
MCP servers for inbox automation
A practical guide to wiring an MCP server to a real inbox. What MCP gives you, what it deliberately leaves out, and where Helix sits relative to a hand-rolled MCP server.
April 29, 2026 · 7 min read
Scoped permissions for AI agents
Most AI agent libraries expose tool functions and stop there. Production agents need scopes — explicit, revocable, audit-able permissions that bind a tool call to an identity. Notes from building Helix.
April 22, 2026 · 6 min read
The approval loop is the product
Most "AI assistant" stories are about the model. The Helix story is about the moment the human says yes. Notes on why the approval loop is the part of the product we polished hardest.
April 15, 2026 · 7 min read
Voice-first inboxes — design notes
Design notes from building a voice-capable email triage loop on top of the Web Speech API. What worked, what failed, and why we kept typed input as a peer of voice — not a fallback.
April 8, 2026 · 5 min read
Why AI agents need their own email address
Borrowing the user’s mailbox for an AI agent is a footgun. A dedicated agent inbox solves audit, scope, and revocation in one move — and lets the human keep their real inbox to themselves.
April 1, 2026 · 6 min read
What is agentic identity?
Agentic identity is the bundle of permissions, addresses, calendars, contacts, and approval rules that an AI agent needs to act on a user’s behalf. Why it matters, and how Helix treats it as a first-class object.
More resources