Set Up Your AI Agent
Connect your AI agent to Hyperstack so it answers from the current documentation, manages your infrastructure, or runs on AI Studio models.
Your AI agent can work with Hyperstack in three ways. Each one is set up separately, so connect only the ones you need.
| What your agent does | What you set up | What it needs |
|---|---|---|
| Answers Hyperstack questions from the current documentation | The Docs MCP Server and the hyperstack agent skill | Nothing: no account and no API key |
| Creates and manages resources on your account | The API MCP Server | You sign in with your Hyperstack account |
| Runs on models from the AI Studio catalog | An AI Studio integration for your tool | An API key |
Start with a prompt
If your agent can edit its own configuration, such as a coding agent, paste this prompt into it. The agent connects the documentation, installs the skill, and confirms that both work. It asks you before it connects anything that acts on your account.
Set up Hyperstack for this agent.
1. Add the Hyperstack Docs MCP server:
- Name: hyperstack-docs
- URL: https://docs.hyperstack.cloud/mcp
- Transport: Streamable HTTP
- Authentication: none
- Setup guide: https://docs.hyperstack.cloud/docs/libraries/docs-mcp/
2. Install the hyperstack agent skill:
- Download https://docs.hyperstack.cloud/.well-known/agent-skills/hyperstack/SKILL.md
- Save it as hyperstack/SKILL.md in this agent's skills directory.
- Skip this step if this agent does not load skills.
3. Confirm the server works:
- Run the search_docs tool for "How do I deploy a virtual machine?"
- If the tool is not available until you restart, tell me to restart you and ask again.
4. Ask me whether to also add the Hyperstack API MCP server:
- URL: https://console.hyperstack.cloud/ai/mcp
- It creates and manages real resources on my account, which can incur charges.
- If I say yes, add it and tell me how to sign in.
- Do not sign in for me.
If you cannot add MCP servers, read https://docs.hyperstack.cloud/llms.txt instead.
To set things up yourself instead, follow the sections below.
Give your agent the documentation
Your agent's training data can be out of date. With the documentation connected, it answers from the current pages and links to each source.
Connect the Docs MCP Server
The Docs MCP Server searches the documentation, retrieves full pages, explains API errors, and looks up GPU flavor specifications. It is read-only and needs no account. Choose your client:
The server URL, for any MCP client (Streamable HTTP, no authentication):
https://docs.hyperstack.cloud/mcp
For each client's configuration file and how to verify the connection, see the Docs MCP Server connection guide.
Install the agent skill
The hyperstack agent skill tells your agent when to use the Docs MCP Server, how to read pages as Markdown when the server is not connected, and which questions need live data from the console or the API instead. The skill is read-only and never acts on your account.
The skill is a single Markdown file. To install it in a client that loads skills from disk, run this command in that client's skills directory. It saves the file as hyperstack/SKILL.md:
mkdir -p hyperstack && curl -fsSL -o hyperstack/SKILL.md https://docs.hyperstack.cloud/.well-known/agent-skills/hyperstack/SKILL.md
Tools that discover skills find it in the skill index at https://docs.hyperstack.cloud/.well-known/agent-skills/index.json. For the index format, Markdown pages, the page index, and the OpenAPI specifications, see Agent-Friendly Docs.
Let your agent manage your infrastructure
The API MCP Server turns plain-English requests into Hyperstack API operations, so your agent can create and manage virtual machines, Kubernetes clusters, and volumes, look up environments, flavors, and GPU stock, and read billing and usage information.
The API MCP Server creates, modifies, and deletes real resources on your account, which can incur charges. Review each action before you confirm it.
Point your client at the server URL:
https://console.hyperstack.cloud/ai/mcp
There is no API key to configure. The first time your client connects, your browser opens the Hyperstack sign-in page, and every tool call then runs against your account. In Claude Code, for example, execute the following command, then run /mcp in a session, select hyperstack-mcp, and complete the sign-in:
claude mcp add --transport http hyperstack-mcp https://console.hyperstack.cloud/ai/mcp
To confirm the connection, ask your agent to list your Hyperstack virtual machines. It returns the virtual machines on your account, or reports that you have none.
For Claude Desktop, Open WebUI, and other clients, see the API MCP Server connection guide. For every tool the server offers, see the tools and operations reference.
Run your agent on AI Studio models
AI Studio serves models through an OpenAI-compatible API, so coding agents and AI tools can use it as their model provider. Each integration guide shows how to point a tool at AI Studio with your API key and choose a model from the catalog.
Claude Code
Run Claude Code on AI Studio models through a local proxy.
Cursor
Use AI Studio models for coding assistance in Cursor.
OpenCode
Run OpenCode on AI Studio models.
RooCode
Run RooCode agents on AI Studio models.
Kilo Code
Run Kilo Code on AI Studio models.
Zed Editor
Use AI Studio models for AI-assisted coding in Zed.
For chat interfaces, gateways, and workflow tools such as Open WebUI, LiteLLM, and n8n, see all AI Studio integrations.
Next steps
Hyperstack MCP Servers
Compare the Docs MCP Server and the API MCP Server.
Agent-Friendly Docs
Markdown pages, the page index, the agent skill, and the OpenAPI specifications.
Getting Started
Select what you want to run and how you want to work, and find the guide that fits.