Skip to main content
The Hyper MCP server brings everything you can do on Hyper — your app integrations, built-in tools, skills, and native platform features — into any AI app that supports the Model Context Protocol. One endpoint, one auth, hundreds of tools. That includes:
  • Your app integrations — Gmail, Shopify, HubSpot, Slack, Notion, Google Drive, and the rest of the 56+ integrations you’ve connected to Hyper
  • Ad platforms — Meta, Google Ads, TikTok, LinkedIn, Pinterest, and others
  • Hyper’s built-in scrapers and generators — Meta Ads Library, Reddit, Twitter, Instagram, TikTok scrapers; image generation; video generation
  • Hyper’s native platform — create and manage agents, tasks, skills, files, knowledge bases, and tables from anywhere
Instead of stitching together dozens of custom MCP servers and managing 40+ API keys, point your AI app at one endpoint and you’ve got the whole Hyper toolbox.

​
Why use Hyper MCP

One endpoint, hundreds of tools

Connect once and get every integration, scraper, generator, and platform tool you’ve enabled on Hyper.

Bring your existing setup

Your connections, agents, tasks, and skills carry over. The work you’ve done on Hyper isn’t trapped on Hyper.

Per-tool guardrails

Enable, disable, and require approval on individual actions across every tool — agents can’t quietly delete a campaign without your sign-off.

Use Hyper from anywhere

Works with Claude, ChatGPT, OpenCode, Cursor, Codex, and any other MCP-compatible AI client.

​
What you can do

Two things, end to end: 1. Use Hyper’s tools inside another AI app. Run scrapers, query your integrations, generate images, browse the web — without leaving Claude, ChatGPT, or wherever you work. 2. Drive the Hyper platform from another AI app. Create agents, assign tasks, attach skills, build tables, and manage files in your Hyper workspace by chatting with Claude or ChatGPT. The agents and tasks you build show up in Hyper, ready to run on schedule or in response to webhooks.

​
How the tools work

Hyper has thousands of tools. An MCP client that receives every one of them sends a very large tool list with each request, and some clients stall on it. So every Hyper MCP server lists the same small set of tools. Everything else is found with search and run with a call tool. A typical run looks like this:
  1. search("meta campaign spend last 7 days") returns meta_ads_campaign_insights_get with call_with: call_read.
  2. describe("meta_ads_campaign_insights_get") returns its input schema.
  3. call_read("meta_ads_campaign_insights_get", {...}) runs it.
Each tool carries an access class: read, write, or destructive. The class decides which call tool runs it, and it decides when your AI client asks you before a call. In Claude, read tools run without a prompt, and write and destructive tools ask you first. A call tool refuses a tool of the wrong class and names the correct one.
Tool names from before this change still work. Pass an old name to the call tool of its class, or call it directly.

​
Two endpoints

Use the endpoint from your MCP page for Cursor, Codex, ChatGPT, custom MCP servers, and any client where you want the full tool set. Use the directory listing in Claude when you want to connect without a custom connector URL. One Hyper login works on both, and your connected apps carry over.

​
Get started

Set up your MCP server

Configure your default server, choose which tools to expose, and set guardrails per integration.

Connect to Claude

Step-by-step walkthrough for connecting Hyper MCP to Claude (works the same for ChatGPT and other clients).