Deploy and debug Zenavo projects from Claude Code

When you finish a feature and want to ship it — or when a service suddenly fails to start — you usually end up bouncing between your terminal and the browser console: finding the project, clicking deploy, waiting on the build, and scrolling through logs.

Connecting Zenavo's MCP server and Skills to Claude Code brings all of that into the same terminal conversation. You can ask: "Check why the latest deployment of web failed, pull the error lines from the logs, and suggest a fix."

If you just want to ask an AI assistant in the browser without installing anything locally, click Zenavo Agent on the right edge of any console page: it sees which page you have open, reads logs and deployments, and proposes changes (deploy, roll back, restart, or set variables) as confirmation cards right in the page. This guide covers wiring the same capabilities into Claude Code on your machine.

1. Install and sign in to zenavo-mcp

Download the binary for your OS from zenavo-mcp Releases, place it on your PATH, and run:

zenavo-mcp login

Your browser opens the device authorization page. Confirm the code matches your terminal, then choose the permission level:

• Read-only: lets the agent inspect projects, service status, deployment history, logs and environment variable names (never plaintext values) — ideal for everyday debugging;
• Deploy: adds the ability to trigger deployments, restart services, roll back to a previous release and update environment variables.

Under project access, choose Selected projects only and check the projects you're working on. Once authorized, the token is saved locally on your machine — you never paste an API key into the chat.

Next, install the Zenavo plugin in Claude Code (which sets up the MCP server plus skills for deploying, debugging, rolling back and managing variables):

claude plugin marketplace add Zenavo-App/zenavo-skills
claude plugin install zenavo@zenavo

If you only want the raw MCP server without the skills, run claude mcp add zenavo -- zenavo-mcp instead. Setup instructions for Cursor, Windsurf and other editors are on the AI Agents docs and the Skills page.

2. Inspect project status and recent deployments

Start with a read-only prompt to verify Claude Code sees your project:

Check my project on Zenavo and list each service's status and latest deployment result.

Claude Code calls the MCP server and lists your services, their current state (Running, Crashed, etc.) and recent build or deploy jobs.

By design, the MCP server never returns plaintext environment variable values (only variable names), and it cannot read database rows or download backups. Even when you share debugging context with the model, your secrets and customer data stay out of the conversation.

3. Pinpoint build or startup failures from the logs

When a deployment fails or a container keeps restarting, let Claude Code inspect the logs by stage instead of hunting through them yourself:

Inspect the latest failed deployment of the web service.
Check the logs to see whether it failed during dependency install, build, container startup or the health check, show the relevant error lines, and suggest a fix.

Once you see the root cause:

• Dependency or build error: have Claude Code fix the code locally, verify the build, commit and push to GitHub;
• Missing environment variable: let it add the variable (after your confirmation) and restart the service;
• Port or health check mismatch: verify the app listens on 0.0.0.0 and on the same HTTP port configured on the service.

4. Trigger a deployment and verify startup

When you're ready to ship:

Deploy the web service in my staging project, follow the build and rollout until it finishes, and check the startup logs for errors.

Remember: Zenavo deploys from the remote Git branch or Docker image configured on the service. If you edited files locally, git push them first before asking Claude Code to deploy.

Any action that mutates state (deploying, restarting, changing variables or rolling back) prompts you to confirm the target service and parameters before running.

5. Roll back quickly when a release has a bug

If a new release misbehaves in production, ask Claude Code to roll back to the previous working deployment:

Check the deployment history for web, find the last successful deployment, roll back to it, and confirm the service is healthy again.

One important caveat: service rollback restores that deployment's image and environment variables, not your database schema or volumes (see Rollbacks). If the broken release also ran a destructive database migration, check that the previous code version is still compatible with the current schema before rolling back.

6. Audit logs and revoking access

Every deploy, restart, rollback and variable change performed through MCP is recorded in your Audit Log.

Keys created by zenavo-mcp login appear under Account → API Keys named MCP on <hostname> and expire after 90 days. If you switch machines or no longer need the connection, delete the key there to revoke access immediately.

FAQ

Can it buy servers, delete projects or read my database tables? No. Zenavo MCP intentionally excludes high-risk actions such as buying or deleting servers, deleting projects, changing billing, opening container shells, reading database rows and downloading backups. Those remain console-only actions.

The plugin is installed, so why can't it see my project? Make sure you ran zenavo-mcp login in your terminal and that the project was checked when you authorized the device. To add more projects, run zenavo-mcp login again.

Should I use Read-only or Deploy permissions? Use Read-only for production projects so Claude Code can inspect logs and status while you review fixes; use Deploy on staging or side projects where you want it to handle the full "fix → push → deploy → verify logs" loop.

How does the built-in Zenavo Agent compare to Claude Code with MCP? Zenavo Agent is built into the web console (included on Dev, Pro and Team plans with 30, 100 and 300 monthly credits) — open any service page, ask why it's restarting, and confirm the proposed fix with one click. Claude Code + zenavo-mcp runs in your local terminal alongside your repository, so it can read production logs, edit your code locally, push to GitHub and redeploy in one loop. If you also want Claude Code to switch across model providers or share team quotas, pair it with a self-hosted Magpie or New API gateway on Zenavo — and connect the same gateway to Open WebUI or LobeHub for browser chat.

Try it on a test project first: run zenavo-mcp login, install zenavo@zenavo, and let Claude Code inspect a log or ship a release.