Skip to main content
The Model Context Protocol (MCP) is a standard for connecting AI tools to external systems. The Helium MCP server gives your AI tool a read-only view of your Helium organization: paywalls, workflows, triggers, targeting, experiments, metrics, and these docs. Ask questions like “which paywall is live on my onboarding trigger?” or “how did trial starts move last week?” and the tool answers from your real dashboard data.
The Helium MCP server is read-only. It cannot edit, publish, or delete anything in your organization.

Connect

The server URL is the same for every client:
There are no API keys to copy. When your client connects for the first time it opens the Helium login page in your browser. Sign in, choose the organization you want to connect, approve the permissions, and you are done. Pick your client: Any other client that supports remote MCP servers with OAuth works the same way: give it the URL above and sign in when prompted.

Permissions

During sign-in your client asks for these permissions. Each one unlocks the tools listed in Available tools below; a tool you have not granted returns a message telling you which permission to add. Metrics and documentation tools are available to every connection. Everything is scoped to the organization you picked at sign-in. To switch organizations, disconnect the server in your client and connect again.

Available tools

Paywalls

Workflows, triggers and targeting

Experiments

Metrics

Docs and account

Good to know

A few things the tools report exactly as the SDK sees them, which can differ from what a paywall’s edit view shows:
  • A paywall’s products, product buckets and content come from its published version. A paywall that has never been published is not served anywhere.
  • A trigger with no workflow attached shows no paywall.
  • Targeting rules are evaluated top to bottom and the first matching rule wins. A rule can point at a paywall, at an experiment, or at “show no paywall”.
  • An experiment serves only while a workflow rule points at it and it is not stopped. Its start and end dates are labels, not a schedule.
  • Variant percentages that do not add up to 100 are scaled so that they do.

Security

  • The server is read-only and can only see the organization you connected. There are no tools that change anything.
  • Your client stores a token for the connection; Helium does not store it. Remove the connection from your client to stop access.
  • Treat tool output like any other data you paste into a chat: paywall names and descriptions are your own content, but review before sharing a transcript outside your team.

Troubleshooting

  • “This tool needs the paywalls:read permission”: reconnect the server and approve all permissions on the consent screen.
  • Sign-in loops or 401 errors: remove the server from your client and add it again; the stored token may belong to an organization you no longer have access to.
  • Tool results are cut off: list tools are paged. Ask for the next page or narrow with a name search.
  • Your client does not show the newest tools: claude.ai and Claude Code cache the tool list. Refresh the Helium server in your client’s connector settings, then start a new chat.