Skip to main content
Your Hyper MCP server is the single endpoint that exposes your Hyper workspace to outside AI apps. You decide which integrations and built-in tools it includes, and you control the permissions on every individual action.

​
Open your MCP page

Go to MCP in the sidebar of your Hyper workspace. Every workspace gets a default server — open it to start configuring.

​
Choose which tools to expose

Your MCP server can include any combination of:
  • App integrations — every app you’ve connected on the Apps page (Gmail, Shopify, HubSpot, Meta, Google Ads, etc.)
  • Built-in tools — Hyper’s native scrapers (Meta Ads Library, Reddit, Twitter, Instagram, TikTok), image generation, video generation, web search, web browsing
  • Platform tools — create and manage agents, tasks, skills, files, knowledge bases, and tables from outside Hyper
Toggle each tool on or off depending on what you want available in the connected AI app. Most users start by enabling everything they actively use.

​
Set per-action guardrails

Click into any integration on your MCP server to see every action it exposes. For each action you can:
  • Enable / disable — turn the action on or off entirely
  • Require approval — the action runs only after you give it the green light
Approval mode is your safety net for destructive or irreversible actions. A common pattern is to enable “send email” but require approval, or enable “create campaign” but require approval on “delete campaign.”
Over MCP, the confirmation happens in your AI client, not in Hyper. Claude asks you before every write or destructive call. A tool you marked “Require approval” shows requires_approval: true in search results, so the client knows to confirm it. Hyper agents and scheduled tasks still pause for approval inside Hyper.
For example, on the Meta Marketing API integration you might:
  • Enable “Get campaign insights” without approval (read-only, safe)
  • Enable “Create campaign” without approval (you want fast iteration)
  • Enable “Delete campaign” with approval (irreversible, you want to sign off)
These guardrails apply to every AI client connected to your Hyper MCP — so the same rules work whether the request comes from Claude, ChatGPT, or anywhere else.

​
Verify your setup

Once configured, your MCP page shows the total number of tools available across the server. A typical fully-configured workspace exposes a few hundred tools — every action across every integration plus the platform tools. Your AI client sees a shorter list: about ten tools, such as search, describe, call_read, and call_write. That is expected. The client finds the hundreds of tools with search and runs them with a call tool. See How the tools work. When you’re happy with the configuration, head to Connect to Claude to wire it up.