How to Manage Cloud Infrastructure With Your AI Assistant Through MCP

Your assistant can explain a cloud configuration. The next useful step is helping you inspect and operate the real resources behind it. That requires an API connection, the right credentials and a way to verify what actually changed.

Short answer: use a cloud provider’s MCP server to expose supported API operations to a compatible AI assistant. Nirvana provides @nirvana-labs/nirvana-mcp. Connect it using the provider’s setup instructions, start with inventory and documentation queries, and check the resulting resource state after each approved change.

TL;DR

• MCP connects an assistant to tools and context; the cloud API still determines available operations and permissions.
• Nirvana’s MCP server supports setup with Cursor, VS Code and Claude Code.
• The current server uses documentation search and a code tool built around the TypeScript SDK.
• Begin with read-only checks, then review proposed resource changes before executing them.
• Cloud API credentials and Agent Sandbox credentials are separate.

What does MCP add to cloud management?

The Model Context Protocol gives AI applications a common way to discover and invoke tools. A host application connects to a server, which exposes capabilities the assistant can use. The useful distinction is between producing a suggested configuration and executing an authenticated API operation.

For a cloud task, three things must agree: your intent, the operation exposed by the server, and the permissions of the credential. A confident answer in chat is not evidence that a VM exists or that a volume has finished expanding. The returned operation and the resource’s observed state provide that evidence.

How does Nirvana’s MCP server work?

The official MCP server README describes a code-mode design. The assistant can search API/SDK documentation and execute code against the TypeScript SDK in a restricted execution environment. That execution environment is part of the MCP tool implementation; it is separate from the Agent Sandboxes you provision for your own workloads.

Start from Nirvana’s SDKs and tools directory and the published MCP package. The README contains client-specific instructions. Record the package version you validate so another engineer can reproduce the setup.

1. Choose the client. Follow the matching Cursor, VS Code or Claude Code configuration instructions. Their configuration locations and secret handling differ.
2. Configure the organization key. The server uses NIRVANA_LABS_API_KEY. Keep the value in the client’s supported secret or environment configuration, rather than in prompts or committed examples.
3. Confirm the connection. Check that the client exposes the server’s documented tools without authentication errors.
4. Inspect before changing. Ask for the available SDK methods and current resources relevant to your task. Confirm the organization, region and resource identifiers.
5. Review a small change. Specify the intended resource, constraints and expected outcome. Review the proposed operation, including any cost or public-network effect.
6. Verify completion. Inspect the operation result and then retrieve the resource. Record its identifier, final configuration and any remaining work.

What should the first prompts look like?

Use bounded requests with an observable finish line. These are illustrative prompts, not claims that every phrase maps to a dedicated tool:

“Find the documented SDK method for listing VMs. Retrieve the current inventory and summarize the names, IDs and regions. Make no changes.”

“Show the API operation and required fields for this proposed VM configuration. Explain the storage and networking choices. Wait for my approval before creating it.”

“Check the operation you just submitted. Retrieve the resource and compare its actual settings with the requested settings. Report anything still pending.”

This progression keeps a useful record of what was requested, what ran, and what exists. It also makes troubleshooting easier: an authentication failure, an unsupported method and an unfinished provisioning operation require different fixes.

Where should MCP stop and Terraform begin?

Use conversational tooling for exploration, inspection and bounded operational tasks. For a repeatable environment that needs review, version history and a reproducible desired state, use an infrastructure-as-code workflow such as Nirvana’s Terraform provider. The assistant can help prepare that configuration without making the chat transcript your only infrastructure record.

For production access, test the client’s approval behavior and the credential’s actual scope. The MCP security guidance is a starting point for the protocol boundary; your cloud account’s authorization and your organization’s change process still apply.

How do you know the setup is ready?

• A read-only request returns the expected organization’s resources.
• An invalid or expired credential produces a clear failure.
• A proposed change identifies its resource, region and expected effect.
• A completed change is checked against the live API response.
• The team can reproduce the configuration without copying secrets from chat.

Frequently asked questions

Does Nirvana have an MCP server?

Yes. The official package is @nirvana-labs/nirvana-mcp, linked from Nirvana’s SDK directory.

Does connecting MCP give the assistant unrestricted cloud access?

Connection alone does not define authorization. Review the credential’s permissions, exposed capabilities and client approval controls before operational use.

Can I use my cloud API key as an Agent Sandbox key?

No. The Agent Sandboxes documentation specifies a separate sandbox credential and header. Verify the correct authentication path for the service you are calling.

Ready to connect your assistant? Start with Nirvana’s MCP and SDK setup. For the execution environment itself, see our code-interpreter backend guide.

Primary documentation reviewed 20 October 2026. This is a setup and verification guide, not a report of a new integration test.

About Nirvana Labs

Nirvana Labs is a high-performance storage cloud built for blockchain, AI and databases: demanding, real-time, stateful workloads. Its high-performance block storage, compute, private networking, Nirvana Kubernetes Service and Agent Sandboxes help teams build and operate data-intensive applications.

Learn more at Nirvana Labs.

Nirvana Cloud | Pricing | Blog | Docs | Changelog | LinkedIn | Twitter | Telegram | YouTube