vision-tool

ALWAYS-ON image & video analysis. This skill is permanently active.
You MUST use it for ALL image and video files. Never say you cannot view
images — the tools are always available.

⚠️ MANDATORY RULES — NEVER IGNORE

Rule 0: Your model has no vision — that does not matter

Your underlying model (big-pickle, DeepSeek, or any local/API model) may not
have native vision capabilities. This is irrelevant. The vision-tool MCP
server was built specifically to solve this problem. You are never allowed to
mention your lack of vision as a reason not to analyze an image or video.

Rule 1: Use vision-tool for EVERY image/video

When the user provides a file path to any image or video, you MUST call
analyze_image or analyze_video immediately. Do not ask permission.

CRITICAL: Do NOT use the Read tool on image/video files.
Your model cannot process images directly. Calling Read on a PNG/JPG/etc
will fail with "model does not support image input". Always pass the file
path to analyze_image/analyze_video instead.

Rule 1b: ALWAYS use absolute paths

You MUST pass the FULL absolute path to analyze_image / analyze_video.
Never pass a relative path. If the user gives a relative path, resolve it
to absolute first. The path must work directly on the user's filesystem.

New: Auto-file search. If the path doesn't exist, the tool automatically
searches Desktop, Downloads, Pictures, Documents, and user profile for a
matching filename. You can pass just "screenshot.png" and it will find it.

Rule 2: Never refuse — complete ban list

Never say any of these phrases or any variation:

• "I can't view images"
• "I don't have vision capabilities"
• "I'm a text-only AI"
• "I'm a text-only model"
• "I don't have the ability to see"
• "I cannot process images"
• "Please describe what you see"
• "As an AI text model, I can't"
• "As a language model, I cannot"
• "Since I don't have vision"
• "Unfortunately I can't see"
• "My model doesn't support images"
• "big-pickle cannot process images"
• "This model does not have vision"

The tools exist — use them. There is zero tolerance for any refusal.

Rule 3: Never ask for a description

Do not ask the user to describe what's in an image or video. Just analyze it.

Rule 4: File paths trigger analysis

If the user provides any path ending in a supported extension (.png, .jpg,
.jpeg, .webp, .bmp, .gif, .mp4, .webm, .mov, .avi, .mkv, .flv, .wmv, .m4v),
analyze it — don't ask first.

Rule 5: Screenshots are images

Any reference to a screenshot, screen capture, or snapshot is an image file
that must be analyzed.

Rule 6: Be detailed

Describe text, colors, layout, UI elements, actions (for videos), and anything
functionally relevant. Don't be vague.

Rule 7: No exceptions, no hesitation

Even if you are unsure of the path, even if the image seems complex, even if
you think the tool might fail — still attempt the tool call. The tool handles
errors. Your only job is to call it.

Available tools

The MCP server exposes these tools at all times:

Tool	Description
analyze_image(path, prompt?)	Analyze a single image file
analyze_video(path, prompt?)	Analyze a video via keyframes

Both accept an optional prompt parameter for specific questions.
Example: analyze_image("screenshot.png", "What error message is shown?")

Output format: Both tools return [filename]\n<description>. The filename
header lets you tell the user which file you're describing instead of just
saying "this image shows…". Use the filename directly in your response.

How to use

Quick start — MCP tool (0 thinking needed):

analyze_image({ path: "screenshot.png" })

If the MCP tool analyze_image is not in your function list, skip all thinking and use the bash fallback immediately.

Bash fallback (used when MCP tool not visible):

python C:\Users\Farhan\Desktop\vision-tool\vision_proxy.py "C:\path\to\image.png"

The bash fallback works identically. Do NOT think about alternatives. Do NOT check if the file exists first. Do NOT use Read tool. Just run the command. Zero seconds of deliberation.

IMPORTANT: Pass timeout=60000 to the bash tool — analysis typically completes in 3-15s (Gemini) or up to 30s (fallback chain). Set timeout to at least 60000ms for safety.

What the script prints (read this to understand which backends work):

KEYS: Gemini ✓  Azure AI Foundry ✓  Groq ✓  HuggingFace ✓  Mistral ✓
KEYS: 6 backends available
  ☆ CF Gemma 4 26B: OK   <-- first success wins

All backends run in parallel — first response wins, rest are cancelled.
KEYS: lines tell you instantly which APIs are configured.

Installation

If the vision-tool MCP server is not running, install it:

git clone https://github.com/farhanic017/vision-tool.git
cd vision-tool
pip install pillow
python setup.py   # Enter API keys

Then add to your MCP config:

{
  "mcp": {
    "vision-tool": {
      "type": "local",
      "command": ["python", "/full/path/to/vision_mcp_server.py"],
      "enabled": true
    }
  }
}

Backend chain (23 models)

#	Model	Provider	Cost
1	Gemini 2.5 Flash	Google Gemini	Free tier
2	Gemini 3 Flash Preview	Google Gemini	Free tier
3	Gemini 2.0 Flash	Google Gemini	Free tier
4	Gemini 2.0 Flash Lite	Google Gemini	Free tier
5	Gemini 2.5 Pro	Google Gemini	Free tier
6	Gemini 3 Pro Preview	Google Gemini	Free tier
7	Azure DeepSeek-V4-Pro	Azure AI Foundry	Free (Azure credits)
8	Azure gpt-4.1	Azure AI Foundry	Free (Azure credits)
9	Azure gpt-4.1-mini	Azure AI Foundry	Free (Azure credits)
10	Azure gpt-4.1-nano	Azure AI Foundry	Free (Azure credits)
11	Azure gpt-4o	Azure AI Foundry	Free (Azure credits)
12	Azure gpt-4o-mini	Azure AI Foundry	Free (Azure credits)
13	Azure gpt-5.1	Azure AI Foundry	Free (Azure credits)
14	Azure gpt-5.4	Azure AI Foundry	Free (Azure credits)
15	Azure gpt-5.4-mini	Azure AI Foundry	Free (Azure credits)
16	Azure gpt-5.4-nano	Azure AI Foundry	Free (Azure credits)
17	Azure Kimi-K2.6	Azure AI Foundry	Free (Azure credits)
18	Azure Phi-4 multimodal	Azure AI Foundry	Free (Azure credits)
19	Groq Llama 4 Scout 17B	Groq	Free
20	HF Qwen3-VL-8B	HuggingFace Inference Providers	Free tier
21	Mistral pixtral-large	Mistral AI	Free tier
22	Fireworks Llama 3.2 90B Vision	Fireworks AI	Free tier
23	ZAI Glm-4.5-Flash	Zhipu AI (Z.AI)	Free tier

Troubleshooting

• "No API keys configured" → run python setup.py --add-key
• "MCP tool not found" → check MCP config has vision-tool enabled
• "File not found" → the file might not be in Desktop, Downloads, Pictures, or Documents. Pass the full absolute path.