MCP server

ComputeSphere ships a hosted Model Context Protocol server. Connect it to your AI assistant once and it can inspect your projects and environments, read status and logs, deploy, scale, restart, roll back, and manage variables — without you copy-pasting curl commands or docs pages.

Serverhttps://mcp.computesphere.com/mcp

Authentication is OAuth — your assistant prompts you to sign in to ComputeSphere on first use, and the connection stays scoped to your user. Revoke it any time from Settings → User tokens.

Add to your assistant

ClaudeAnthropic

Recommended

Add to Claude→Manual setup↗

CursorAnysphere

Add to Cursor→Manual setup↗

ChatGPTOpenAI

Manual

Manual setup↗

For Claude Desktop, Windsurf, Zed, VS Code, or Continue — copy the JSON config below into your client's MCP settings file.

Manual setup (any MCP client)

ComputeSphere speaks streamable HTTP (the standard MCP transport for hosted servers). Drop this into your client's MCP config:

{
  "mcpServers": {
    "computesphere": {
      "url": "https://mcp.computesphere.com/mcp"
    }
  }
}

Client	Config file location
Claude Desktop	~/Library/Application Support/Claude/claude_desktop_config.json (macOS) · %APPDATA%\Claude\claude_desktop_config.json (Windows)
Cursor	~/.cursor/mcp.json
Windsurf	~/.codeium/windsurf/mcp_config.json
Zed	~/.config/zed/settings.json (under "context_servers")
VS Code	Workspace .vscode/mcp.json or via code --add-mcp (uses "servers" instead of "mcpServers" — see below)
Continue	~/.continue/config.json (under "mcpServers")

VS Code names the top-level key "servers" and takes the transport type:

{
  "servers": {
    "computesphere": {
      "type": "http",
      "url": "https://mcp.computesphere.com/mcp"
    }
  }
}

Restart your client after editing. The connection authenticates on first tool call via your browser.

What your assistant can do

Once connected, ComputeSphere exposes its tools to the assistant. You don't call them directly — you describe what you want and the assistant picks the tools. The tools mirror the ComputeSphere API, so each one needs the same project role the API does: Viewable to read, Editable to change things, Admin to read variables or delete.

These are the ones you'll use most:

Tool	What it does	Project role
listProjects	List the projects in an account.	Viewable
listEnvironments	List the environments in a project.	Viewable
listServices	List the services in a project.	Viewable
listDeployments	List deployments.	Viewable
getDeploymentStatus	Read a deployment's current status.	Viewable
getDeploymentRuntimeLogs	Read the runtime logs of a deployment's spherelets.	Viewable
getDeploymentDeployLogs	Read a deployment's rollout logs.	Viewable
getDeploymentBuildLogs	Read the build logs for one version.	Viewable
getDeploymentMetrics	Read CPU and memory metrics.	Viewable
createDeployment	Create a deployment.	Editable
deployDeployment	Start or re-issue a deployment.	Editable
applyManifest	Apply a computesphere.yaml manifest. Asks you to confirm.	Editable
scaleDeployment	Change the spherelet count.	Editable
restartDeployment	Restart a running deployment.	Editable
rollbackDeployment	Roll back to an earlier version.	Editable
setDeploymentEnvVar	Set one variable.	Editable
listDeploymentEnvVars	Read a deployment's variables. Secret values are never returned, only their names.	Admin
deleteDeployment	Delete a deployment. Asks you to confirm.	Admin

That's a selection, not the full set: the server also covers projects, environments, domains, builds, snapshots, alerts and more. Your assistant receives the complete, current list when it connects, so no client-side updates are needed.

Confirmation for destructive actions. Deleting anything, suspending a deployment, applying a manifest, or replacing a whole set of variables or secrets goes through a confirmation step: the assistant shows you the exact action and it runs only after you approve it.

Try it

Once connected, try one of these prompts:

• "Deploy a hello-world Go service in a new project."
• "Show me the last 100 log lines from my api service in production."
• "Scale worker to 3 spherelets."
• "What's failing in my deployment? Read the deploy logs and tell me."
• "Roll api in staging back to the previous version."

The assistant chooses which tools to call, authenticates if needed, and reports back.

Privacy and scope

• OAuth-scoped. The connection acts as you. It sees the same accounts, projects, and services your user account does. Same RBAC rules apply.
• Revocable. Disconnect any time from Settings → User tokens in the console.
• Secrets stay out of the conversation. listDeploymentEnvVars returns values for plain variables, but only the names of secrets — secret values are never returned, so they never enter assistant context.
• No training on your data. ComputeSphere doesn't log or train on the contents of tool calls or responses. Your assistant vendor's data policy still applies — see Claude's data policy or OpenAI's.

What's next

• Quickstart — ship your first service via MCP, CLI, or console.
• API keys — for non-MCP automation (CLI, SDKs, raw API).
• Pricing — MCP is included on every plan.