How to deploy a Claude Code app to a team URL

You can run the host's deployment tools yourself, connect a provider-specific plugin, or ask SpringRoll to manage the handoff. Direct hosting fits a developer who wants to maintain the provider account, project configuration, credentials and deployment history. A host's plugin brings those provider actions into the coding session. Check the provider's documentation for its setup and access model.

SpringRoll adds a workspace application record, an accountable owner, immutable releases and a launch path through the App Portal. It manages the Vercel or Cloudflare hosting account and chooses among eligible runtimes. This is useful when the next user is a colleague rather than the person who wrote the app. It still needs an application that fits the selected runtime, a working build and appropriate authentication at the origin.

• See the deployment handoff →
• Compare the Retool approach →

Run the command below in your terminal. It registers SpringRoll as a remote HTTP MCP server, using the public endpoint from the connection catalog. You do not need to copy a SpringRoll agent token into your project. Open Claude Code in the folder that holds the app so it can inspect the actual source and build configuration.

Keep the MCP connection and the optional SpringRoll skill distinct. The connection makes the tools discoverable; installing the skill adds the guided deployment workflow and its slash command. A plain-language deployment request works after manual MCP setup. Do not assume that adding a server also installed a skill or changed the application's source.

claude mcp add --transport http springroll https://springroll.dev/api/mcp

• Review all connection choices →

Start Claude Code and enter /mcp, the authorization command shown in the connection catalog. Follow the browser flow, sign in and choose the workspace that should own the application. Review the organization and requested access before accepting. If the browser shows the wrong account, correct the account selection before continuing the coding session.

The authorization acts within one member's role. It does not give Claude every administrator permission, and it cannot record a human approval. A staged release, a new data grant or a change that requires a person's decision still waits for that person. Ask Claude to report the required next action instead of repeatedly submitting a blocked request.

/mcp

• Understand the human review steps →

Use the deployment prompt below after Claude has checked the local project. Supply the real app name, accountable owner and support contact when asked. Have it show the build command, output directory, framework and external services. These details describe what will actually run; a successful development server alone does not establish a production build.

Before it starts, inspect the selected workflow and target environment. Direct to Production is the default. A Tenant Admin can configure Staged when the first release needs Development, UAT and Production review. Do not infer that a preview is the default just because the work began in a coding agent.

The result should identify the application, release and deployment, together with its actual status. A queued build is still in progress. Keep those references and follow the deployment status until it reaches READY or a terminal failure. Read the build error when it fails; an agent saying that it submitted the job is not evidence that a URL works.

Prepare this project for SpringRoll. Check the build, workspace, owner and deployment workflow; tell me the target environment, then deploy and follow the build to its actual result.

Open the returned working URL after READY and perform the app's main task. Compare a visible result with the expected fixture, check a narrow screen and try the empty or error state. Confirm the environment and release before copying the SpringRoll app link. For a staged app, Test and Production serve different handoff needs.

Choose the intended audience and test the portal link with an allowed colleague and a separate account outside that audience. Check the provider URL independently using application or provider authentication. Record the owner, test results and support route so the colleague receiving the link can use the app without access to your terminal session.

• Complete the team handoff checklist →
• Check the direct access boundary →

Search for localhost addresses: a colleague's browser cannot reach the development server on your laptop. Replace development endpoints with configured production services. Check missing environment variables against the deployed environment, and distinguish server-only secrets from values that the framework embeds in browser JavaScript.

A SQLite file or an upload written beside the app is not automatically shared durable storage. Move team records to a suitable database or storage service and test two users. Review background workers and long-lived connections against the selected runtime. Finally inspect the built client bundle for credentials; deleting a secret from the latest source does not revoke an already exposed key.

• Check runtime fit and storage needs →

Use /mcp to inspect the connection from the same Claude Code session that will deploy. If SpringRoll is absent, verify that the add command completed and that the session is reading the configuration you intended. The local CLI help documents local, user and project scopes; choose scope deliberately rather than copying a teammate's private configuration file.

If discovery works but a workspace tool asks you to sign in, complete the browser authorization. If it reports missing permission, ask the workspace administrator to review your actual role. Installing the optional skill does not increase that role. A skill-specific slash command being unavailable is different from a disconnected MCP server: use the plain-language prompt or install the skill from the documented setup page.

• Claude Code MCP documentation →
• Using Codex instead? →
• Follow the worked deployment example →