CLI and MCP server for Google Flow — drive Veo text-to-video, image-to-video and Imagen image generation from the terminal.
The server provides 8 tools with mostly consistent naming (verb_noun pattern) and basic descriptions. However, descriptions are sparse (most under 100 chars, well below the 194-char production baseline), input parameter descriptions are minimal or missing, and no output schemas are documented. Schema definitions show type information but lack depth and validation constraints. No error handling guidance visible. Tool names are reasonably clear but descriptions do not consistently explain WHEN to use each tool or WHAT it returns, failing the pattern:tool-description standard.
Download generated media (video or image) by asset UUID
Generate image using Google Flow Imagen API
Generate video from text or image using Google Flow Veo API
Get details of a specific project
Get details of a specific generation task
Inspect credit balance for one or all profiles
List projects in the current profile's datastore
Descriptions are uniformly short and lack actionable context. 'Generate video from text or image using Google Flow Veo API' (60 chars) is below the 194-char production baseline and does not explain WHEN to call this vs generate_image, WHAT profiles are available, or WHAT the return structure contains.
Input parameter descriptions are minimal. The 'profile' parameter appears in all 8 tools with identical description ('Profile name to use for generation (default: auto-resolve)'), but the description does not explain what profiles are available, how to discover them, or what happens on auto-resolve.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
List generation tasks from queue
No output schemas are documented. The code does not show what fields are returned by each tool or their types. LLMs cannot infer downstream tool usage or data extraction without documented return types.
Parameter constraints are not enforced in descriptions. For example, 'duration' in gflow_generate_video accepts '5, 10, or 20' but this constraint is buried in a generic description and not formalized as an enum. This invites invalid LLM calls.
The 'tools' parameter in gflow_generate_video and gflow_generate_image has type 'array' but no item schema is specified. What is the structure of each tool specification? What are 'name' and 'options'?
No error handling guidance visible in tool descriptions. LLMs do not know what errors are retryable, what indicates a user-fixable issue, or how to recover from failures.