Solutions

Deploy apps built with Claude Code or Cursor to your company cloud

To deploy an app built with Claude Code or Cursor inside a company, the code has to be built from saved source, checked, approved and run somewhere IT controls, with company sign-in in front of it. Diakrin connects to the coding agent through MCP or its CLI, builds and scans the app in the company's own AWS or Google Cloud account, and deploys it only through an approved release plan, with an independent reviewer where your policy requires one.

Last reviewed

To deploy an app built with Claude Code or Cursor inside a company, the code has to be built from saved source, checked, approved and run somewhere IT controls, with company sign-in in front of it. Diakrin provides that path without asking you to leave your coding tool. The agent connects through MCP or the Diakrin CLI, and the app is built, checked and run in your company's own AWS or Google Cloud account once the release is approved under your approval policy.

Why is deploying an AI-built app at work harder than at home?

Most deployment guides for Claude Code or Cursor apps assume a personal project: write a Dockerfile, push to a hosting service, set some secrets and share the link. At work that link becomes a problem. The app needs company sign-in, a database the company owns, credentials that never sit in a prompt, a release someone approved and a cost someone is accountable for. Diakrin handles those parts so the builder only describes and builds the app.

How does a coding agent connect to Diakrin?

Through the Model Context Protocol (MCP ↗) or the CLI. You create one personal connection, not one per app. It acts as you: it reaches every app you can access, and every call is checked against your current role and app access.

ToolHow it connects
Claude CodeAdd Diakrin as a remote MCP server
CursorAdd Diakrin in Cursor's MCP settings
CodexUse the TOML configuration from the console
OpenCodeAdd Diakrin as a remote MCP server in opencode.json
Any terminal or agentThe diakrin CLI, with --json for agents

On BYOC you create a connection token in Settings → AI tools and store it in the tool's local secret environment as DIAKRIN_MCP_TOKEN. The console shows the exact configuration for your tool. See Connect your coding tool.

What can the agent do through MCP?

The agent can read the app's requirements and the trusted developer contract, request data access, save versions, start and follow builds, prepare a release plan, and publish the latest version or roll back. Tools appear only when your current role permits them. The agent cannot approve its own requests, reach another organization or delete apps. The MCP tool reference lists every tool by task.

What happens between save and production?

StepWhoWhat it proves
Save a versionBuilder or agentSource history is recorded. Nothing is published.
CheckYour cloudThe version is built and started with a test database.
PreviewYour cloudA private, expiring preview of that exact version is ready.
Approve the planIndependent reviewerThe exact image, target, cost estimate and access implications were approved.
PublishBuilderAn approved rolling update replaces the old version, stopped by health checks if the new one is unhealthy.
Verify liveDiakrinFresh, server-verified evidence matches the published image.

A successful build is not a preview, and a preview is not production approval. See Preview, approve and deploy.

How it works with Diakrin

  1. Connect your tool once. Create the connection and confirm the agent lists the apps you can access.
  2. Pick or create the app. Ask the agent to select the exact Diakrin app, read its requirements and list any missing prerequisites. For example: "Connect to the Team request tracker app and explain its approved capabilities before building."
  3. Build on the company template. The agent reads the trusted developer contract and writes the app, starting with synthetic data.
  4. Request data access, do not paste it. If the app needs company data, the agent requests a named operation. IT approves it, and an authorized owner enters credentials on a customer-hosted form.
  5. Save and preview. diakrin save or the agent's save tool records the version and runs the checks. Before the first publish it goes to your private preview; later versions show Ready to publish. See Install and use the Diakrin CLI.
  6. Get approval and publish. Where your policy requires it, a reviewer approves the plan in the console, then you publish. Employees open the app with company sign-in.

What if the deployment is refused?

Every release plan is checked against company rules: approved capabilities and data grants, the app's budget ceiling, the reviewed template and the approval policy. A failed rule blocks the release with a reason the builder and agent can act on, or names who must act, such as IT. Nothing is built or deployed on a refusal. For example, when the app's monthly budget is used up, the request is refused with BUDGET_EXHAUSTED until the ceiling is raised or held amounts settle. See Troubleshooting and help.

For the controls IT applies at the approval step, read AI app governance and IT approval.

Frequently asked questions

Which coding tools can deploy through Diakrin?

Claude Code, Cursor, Codex and OpenCode have documented setups. Other tools work if they support MCP over Streamable HTTP with a secure bearer-token configuration. The Diakrin CLI gives terminals and agents the same capabilities.

Do I need to write a Dockerfile or cloud configuration?

No. Apps are built on a reviewed company template. The Diakrin runner builds an immutable image from an exact source commit and a pinned template inside your cloud account, and deploys it to the company cloud setup IT created.

Does saving the app publish it?

No. Saving records a version and starts its checks. Before the first publish, the version is built, checked and taken to a private preview; after that, a checked version shows Ready to publish. Publishing to the company is a separate step that needs approval.

Can I deploy an existing app I already built?

When your company enables source access, you can bring an app from your coding tool or from a company repository. Imported code is assessed without running it, and any adaptation work stays separate from the original.

Where does my MCP token go?

In the coding tool's local secret environment, as DIAKRIN_MCP_TOKEN. Keep it out of chat, project files and source control. Claude Code and Codex on a self-hosted installation sign in through the browser instead, so there is no token to store.

Can my coding agent approve its own release?

No. The agent acts as you, and requesters cannot approve their own plans. An eligible reviewer who is not the requester approves or denies the exact plan.

Sources

Facts about other products come from their public pages, checked on the dates shown. Products change; tell us if something is out of date.

See Diakrin with your own cloud

Let your team build internal apps with the AI tools they already use, while IT keeps control of identity, data, approvals and spend.