Skip to main content
Glama
Qwen3-VL Video Understanding MCP Server

Qwen3-VL Video Understanding MCP Server

by adamanz

Qwen3-VL Video Understanding MCP Server (Blaxel)

An MCP (Model Context Protocol) server that enables Claude and other AI agents to analyze videos and images using Qwen3-VL-8B-Instruct deployed on Blaxel.

Features

  • Video Analysis: Analyze videos via URL with custom prompts

  • Image Analysis: Analyze images via URL

  • Video Summarization: Generate summaries in different styles

  • Text Extraction: Extract on-screen text and transcribe speech

  • Video Q&A: Ask specific questions about video content

  • H100 GPUs: Fast inference on NVIDIA H100 GPUs via Blaxel

Related MCP server: Vision MCP Server

Architecture

Claude/Agent → MCP Server → Blaxel API → Qwen3-VL (H100 GPUs)

Prerequisites

  1. Blaxel Account: Sign up at blaxel.ai

  2. Blaxel CLI: Install the Blaxel CLI

  3. ffmpeg: Required for video frame extraction

  4. Python 3.10+

Quick Start

1. Deploy the Model to Blaxel

cat << 'EOF' | blaxel apply -f -
apiVersion: blaxel.ai/v1alpha1
kind: Model
metadata:
  name: qwen-qwen3-vl-8b-instruct
  displayName: Qwen/Qwen3-VL-8B-Instruct
spec:
  enabled: true
  policies: []
  flavors:
    - name: nvidia-h100/x4
      type: gpu
  runtime:
    model: Qwen/Qwen3-VL-8B-Instruct
    type: hf_private_endpoint
    image: ''
    args: []
    endpointName: qwenqwen3-vl-8b-instruct-nvidia-h100
    organization: adamanz
  integrationConnections:
    - huggingface-4s2m2h
EOF

Or use the provided config:

blaxel apply -f blaxel-model.yaml

2. Get Your API Key

blaxel auth token

3. Install the MCP Server

cd qwen-video-blaxel-mcp
pip install -e .

Or with uv:

uv pip install -e .

4. Configure Environment

cp .env.example .env
# Edit .env with your Blaxel API key

5. Add to Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "qwen3-video-blaxel": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/qwen-video-blaxel-mcp",
        "run",
        "server.py"
      ],
      "env": {
        "BLAXEL_API_KEY": "your-blaxel-api-key",
        "BLAXEL_MODEL": "qwen-qwen3-vl-8b-instruct"
      }
    }
  }
}

6. Restart Claude Desktop

The qwen3-video-blaxel tools should now be available.

Available Tools

analyze_video

Analyze a video with a custom prompt.

analyze_video(
  video_url="https://example.com/video.mp4",
  question="What happens in this video?",
  max_frames=8
)

analyze_image

Analyze an image with a custom prompt.

analyze_image(
  image_url="https://example.com/image.jpg",
  question="Describe this image"
)

summarize_video

Generate a video summary.

summarize_video(
  video_url="https://example.com/video.mp4",
  style="detailed"  # brief, standard, or detailed
)

video_qa

Ask specific questions about a video.

video_qa(
  video_url="https://example.com/video.mp4",
  question="How many people appear?"
)

extract_video_text

Extract text and transcribe speech.

extract_video_text(
  video_url="https://example.com/presentation.mp4"
)

check_configuration

Check the Blaxel API configuration.

list_capabilities

List all server capabilities.

Configuration

Environment Variable

Description

Default

BLAXEL_API_KEY

Your Blaxel API key

Required

BLAXEL_API_URL

Blaxel API URL

https://api.blaxel.ai/v1

BLAXEL_MODEL

Model name

qwen-qwen3-vl-8b-instruct

Requirements

  • ffmpeg: Required for video frame extraction

    # macOS
    brew install ffmpeg
    
    # Ubuntu/Debian
    apt install ffmpeg

Supported Formats

Video: mp4, webm, mov, avi

Image: jpg, jpeg, png, gif, webp

Comparison: Modal vs Blaxel

Feature

Modal

Blaxel

Model

Qwen2.5-VL-7B

Qwen3-VL-8B

GPU

A100

H100

Pricing

Pay-per-second

Subscription

Cold Start

~30-60s

Faster

Setup

Deploy code

Apply YAML

License

MIT

Available Tools

7 tools
analyze_imageA
Analyze an image using Qwen3-VL-8B vision-language model on Blaxel.

The image must be accessible via a public URL.

Examples:
- "What's in this image?"
- "Describe the scene"
- "What text is visible?"
- "Identify any people or objects"

Inputs: image_url (required), question, max_tokens.

analyze_videoA
Analyze a video using Qwen3-VL-8B vision-language model on Blaxel.

The video must be accessible via a public URL. The model will:
1. Download the video
2. Extract key frames (up to max_frames)
3. Analyze the frames with your question

Examples:
- "What happens in this video?"
- "Summarize the main events"
- "What products are shown?"
- "Describe the people and their actions"

Inputs: video_url (required), question, max_frames, max_tokens.

check_configurationC

Check the Blaxel API configuration.

Inputs: no named parameters.

extract_video_textC

Extract text and transcribe speech from a video.

Inputs: video_url (required).

list_capabilitiesB

List all server capabilities.

Inputs: no named parameters.

summarize_videoC
Generate a summary of a video.

Styles:
- brief: 1-2 sentence overview
- standard: 1-2 paragraph summary
- detailed: Comprehensive analysis

Inputs: video_url (required), style.

video_qaB
Ask a specific question about a video's content.

Examples:
- "How many people appear?"
- "What color is the car?"
- "What is being demonstrated?"

Inputs: video_url (required), question (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 7 tools

    Disambiguation3/5

    The tools have overlapping purposes that could cause confusion. analyze_video, summarize_video, and video_qa all process videos with similar capabilities, while analyze_image stands alone for images. The descriptions help differentiate use cases, but an agent might struggle to choose between summarize_video and video_qa for general video queries.

    Naming Consistency4/5

    Most tools follow a consistent verb_noun pattern (analyze_image, analyze_video, extract_video_text, summarize_video, video_qa). However, check_configuration and list_capabilities deviate slightly with different verb styles, though they remain readable and understandable.

    Tool Count5/5

    With 7 tools, the count is well-scoped for a video understanding server. Each tool has a distinct role, covering image analysis, video analysis, configuration checks, text extraction, and capability listing, making the set comprehensive without being overwhelming.

    Completeness4/5

    The tool surface covers core video and image analysis tasks well, including configuration and capability checks. Minor gaps exist, such as no explicit tool for editing or modifying media, but agents can work around this with the provided analysis and summarization tools for typical use cases.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers