Deploy apps with an AI agent over MCP

Use a Resilis account with access to the organization you intend to manage. The MCP endpoint works without a paid plan. Authentication, organization membership, operation permissions, plan limits, and resource billing still apply. Connecting a client does not grant free compute or authorize it to create billable resources on your behalf.

Follow the Codex CLI or Claude Code setup guide to register the remote HTTP server and start OAuth sign-in. Review the requested access in the browser and select an organization if prompted. Check the configured server and available tools before requesting an operation. A server-list entry confirms configuration, not a healthy deployment.

• Codex MCP add server setup →
• Claude Code MCP setup →

For local workspace deployment, the agent needs local file access, command access, and Git. Ask it to inspect the application and propose source, build settings, required variables, and a target. Review tracked, untracked, and ignored files before upload. Keep credentials out of source and configure sensitive values through the supported environment settings.

The workspace route prepares a reviewed commit and transfers source to private Resilis-hosted Git. It verifies the selected commit before deployment. GitHub is not required for this route, but Git and source transfer are. Named build contexts are not supported. Your authorization should identify the project, environment, and compute target as well as the app.

Resilis also supports GitHub source inspection and app deployment. Keep that route distinct from local workspace transfer: the dashboard requires a connected GitHub repository, while the workspace path starts with local files. Use the getting-started documentation to choose the route before asking the agent to execute a plan.

• Deploy your first app →
• GitHub dashboard deployment →

Supported tools cover application deployment and status, configuration, build and runtime log inspection, and metrics for supported resources. Agents can request restart, redeploy, or rollback where the target supports them. Review resource identity and environment for each request, especially when similarly named apps exist in development and production.

Data tools support PostgreSQL and MySQL creation, database binding, and supported protection workflows. Storage tools cover bucket and object operations and external storage connections. These operations require appropriate account access and remain subject to resource capabilities and plan checks. Creating data resources and deleting them are separate authorizations.

Environment configuration includes sensitive variables and custom domains for supported managed applications. Custom domains are unavailable on connected compute and S3-backed static sites. The presence of a domain tool does not remove that restriction. S3-backed static sites also reject runtime log requests because they are not running application containers.

• App operations and connected compute →
• Managed data and recovery →

Resilis is not the coding agent. It does not write your application source simply because an MCP connection exists. The external agent edits files and requests cloud operations. Client file permissions and command permissions are separate from Resilis account access; verify both before expecting a workspace workflow to succeed.

Agent access does not make every production action automatic or reversible. App rollback does not restore data or undo database migrations. Connecting your own VPS or Kubernetes cluster does not enable all managed-cloud features there. Target health, advertised capabilities, and operation-specific restrictions still determine what is available.

Ask the agent for a plan before a deployment, configuration change, or data operation. Check the selected organization, resource, environment, source revision, and intended effect. Authorize only the actions you want performed. Some destructive and production operations have additional action-specific protection; an existing OAuth session is not blanket permission.

If a request is refused, check account membership, selected organization, requested scope, resource capability, and any required confirmation. Do not ask the agent to bypass a permission or production safeguard. Use the client guide for sign-in troubleshooting and contact support with a sanitized error if you cannot identify the cause.

After deployment, ask for status and a bounded log inspection. Open the serving URL. Perform an application action yourself. If data must persist, verify it after replacing the app process. Keep a distinction between what the agent requested, what Resilis reported, and what you observed in the application.

Before creating resources, review usage rates and plan credit. Compute, persistent disk, database fees, egress, and object storage draw from runtime credit; build overage is separate. The billing example shows one assumed workload and excludes transfer and object storage. Use your actual usage when estimating an ongoing bill.

• Usage rates and worked example →
• Compare plans →