cloudeval login or cloudeval login --headless. Use --machine only when service-principal machine authentication is configured.

The exported image should match the same architecture or dependency graph users inspect in the browser.
Why it matters
AI agents and CI jobs need deterministic files, not screenshots from an interactive tab. The CLI keeps the export workflow authenticated and repeatable without requiring agents to drive a browser.Quick example
Download the private architecture and dependency diagrams for one project:diagram-downloads/architecture.pngdiagram-downloads/dependency.png- human output,
--jsondata.output,data.headersOutput, andfilesWrittenreport resolved absolute filesystem paths - optional headers output when you pass
--headers-output
CLI shape
cloudeval login,
cloudeval login --headless, or --machine. The frontend defaults to
https://cloudeval.ai; pass --frontend-url only for local/dev frontends. Use
--public only for intentionally public or shared graph bytes.
Private exports
Private CLI exports requirecloudeval login, cloudeval login --headless, or --machine. Prefer stored login state for user workflows.
PNG with all labels:
Public or shared exports
Use--public only when you intentionally want public or shared graph data. It
is a scope reduction: CloudEval uses only public graph data and never falls back
to private project data.
Validate export metadata
When you pass--headers-output, keep the metadata file next to the image:
MCP tool
The CloudEval MCP server exposes the same workflow asprojects.exportDiagram. It writes the image to outputPath and can optionally write response headers to headersOutputPath.
Security checks
These checks should be part of automation:- run
cloudeval auth status --format jsonbefore private exports - use
--publiconly for diagrams that are already safe to share - keep private and public/share outputs in separate directories
- fail the job when the expected image file is missing or empty
- avoid raw HTTP calls; use the CLI or MCP tool so auth and scope handling stay consistent
Common mistakes
Tips
- Prefer
labels=allfor agent-readable diagrams. - Save headers next to each image so agents can verify provenance later.
- Use CLI JSON output when scripts need structured success or failure details.
- Use the web app for visual review when humans need to inspect layout details.