agent-vision
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| OPENAI_MODEL | No | Compatibility fallback for AGENT_VISION_MODEL. | |
| VISION_MODEL | No | Compatibility fallback for AGENT_VISION_MODEL. | |
| OPENAI_API_KEY | No | Compatibility fallback for AGENT_VISION_API_KEY. | |
| VISION_API_KEY | No | Compatibility fallback for AGENT_VISION_API_KEY. | |
| OPENAI_BASE_URL | No | Compatibility fallback for AGENT_VISION_BASE_URL. | |
| VISION_BASE_URL | No | Compatibility fallback for AGENT_VISION_BASE_URL. | |
| AGENT_VISION_MODEL | No | The model name to use for vision analysis. | |
| AGENT_VISION_API_KEY | No | Your API key for the vision API. Optional if the endpoint has no authentication. | |
| AGENT_VISION_HEADERS | No | Additional request headers as a JSON string. | {} |
| AGENT_VISION_BASE_URL | No | The base URL of your OpenAI-compatible vision API endpoint. | |
| AGENT_VISION_MAX_TOKENS | No | Maximum output tokens. | 4096 |
| AGENT_VISION_TIMEOUT_MS | No | Download and API timeout in milliseconds. | 120000 |
| AGENT_VISION_FFMPEG_PATH | No | Path to ffmpeg executable. If not set, the server uses PATH then the bundled ffmpeg. | |
| AGENT_VISION_MAX_IMAGE_MB | No | Maximum image size in megabytes. | 20 |
| AGENT_VISION_MAX_VIDEO_MB | No | Maximum video size in megabytes. | 200 |
| AGENT_VISION_ALLOW_PRIVATE_URLS | No | Whether to allow private network URLs for images/videos. | false |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| analyze_imageA | Analyze an image from a local path, HTTP(S) URL, or base64 data URI. Use ONLY when you cannot view the image yourself (no native multimodal/image input) — if you can see the image directly, answer from your own vision instead. For screenshots, UI, OCR, diagrams, photos, and visual debugging. |
| analyze_videoA | Analyze a local or remote video by extracting evenly spaced keyframes. Use this for screen recordings, UI flows, demos, and event summaries. |
| generate_imageA | Generate one image from a text prompt using configured OpenAI-compatible or Volcengine Ark (Seedream) providers. Uses the same settings as the standalone CLI. Returns image content and metadata; optionally saves to a new local file. This is image creation, not image analysis, and works regardless of native vision capability. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 3 tools
The three tools target clearly distinct actions and modalities: image generation, image analysis, and video analysis. The analyze_image description also explicitly clarifies when not to use it, reducing misselection risk.
All tools follow a consistent verb_noun pattern: generate_image, analyze_image, analyze_video. The convention is predictable and directly reflects each tool's purpose.
Three tools are well-scoped for a focused vision server: generation, image understanding, and video understanding. Each tool has a clear reason to exist, with no redundant surface area.
The core lifecycle for vision tasks is covered: create images, inspect images, and inspect videos. Minor gaps exist, such as image editing/transformation or batch analysis, but agents can work around them for most stated use cases.