HyperAI MCP
Manage HyperAI compute containers, model servings, datasets, and models from AI tools like Claude Code and Cursor through the Model Context Protocol.
Model Context Protocol (MCP) is an open standard that lets AI applications securely connect to external services. Once an AI tool is connected to an MCP server, it can call the server's tools on your behalf — reading data, running operations, and reporting results back in the conversation.
HyperAI provides an official MCP server so that AI tools such as Claude Code and Cursor can operate the HyperAI platform directly: list and inspect your compute containers, create or stop workspaces, deploy models as inference servings, search datasets and models, and more — all through natural language.
The HyperAI MCP server
The HyperAI MCP server is a remote server available at:
https://mcp.hyper.aiIt uses the Streamable HTTP transport, so there is nothing to install locally — you add the URL to your MCP client, sign in with your HyperAI account in the browser, and start using it. See Connect to HyperAI MCP for client-specific setup.
What you can do
The server exposes tools in six feature groups:
| Group | What it covers |
|---|---|
user | Query your account profile, quota and limits, billing transactions, resource usage, and subscriptions; manage personal access tokens |
compute | List, inspect, create, stop, restart, and delete compute containers — under your own account or an organization via the username parameter; create, update, and delete projects, change their visibility, and manage port mappings and idle timeouts; read container metrics, README, and notebooks |
resources | Search public projects (tutorials and community projects) |
dataset | Search and inspect datasets and models, both your own and public ones; create entries and versions, update metadata, change visibility, and delete them |
org | List the organizations you belong to, view organization details, seat quota, and resource limits, and list members |
serving | Deploy models as inference servings and manage their versions — create, roll out new versions, restart, scale, stop, and delete; read status, resource metrics, and request telemetry; manage access control and API keys |
For the full list of tools and their parameters, see the MCP Tools Reference.
Authentication
The server uses OAuth 2.1 authorization. When you connect for the first time, your MCP client opens a browser window where you sign in with your HyperAI account and approve the connection. After that, every tool call runs under your own account — the AI can only see and operate the containers, projects, and datasets your account has access to. The sign-in also issues a refresh token, so the client keeps the session alive in the background — you won't be asked to sign in again under normal use.
Note
Actions performed through MCP are real operations on your account. Creating a container consumes your compute quota and balance exactly as if you had created it in the console. Before creating or restarting a container, creating a project or a dataset version, changing who can see a project or dataset, deploying or restarting a serving, issuing a personal access token or a serving API key, or deleting a dataset, a serving, a project, or a container run, the assistant restates what it is about to do and waits for your explicit go-ahead — see Confirmation before writes.
Enabling only some tool groups
By default all six feature groups are enabled. If you only need a subset, append a features query parameter to the server URL when registering it:
https://mcp.hyper.ai/?features=compute,datasetUnknown group names are ignored. This is useful for keeping the tool list small in clients where too many tools dilute the AI's attention.