Video Analytics MCP#

The Video Analytics MCP (Model Context Protocol) Server provides AI agents with access to video analytics data through the MCP protocol. It enables natural language querying of incidents, sensor information, and metrics stored in Elasticsearch.

Overview#

The VA-MCP Server is built on NVIDIA Agent Toolkit (NAT) and exposes video analytics capabilities as MCP tools that can be consumed by AI agents.

Key Features:

  • Query incidents by sensor, place, time range, and VLM verdict

  • Semantic place search using sentence-transformer embeddings

  • Field-of-view occupancy histograms and object counts

  • Statistical analysis (average speeds, incident counts)

  • Integration with VST for active sensor filtering

Available Tools:

Tool

Description

get_incident

Retrieve a specific incident by ID

get_incidents

Query incidents with filtering by sensor, place, time range

get_sensor_ids

Get list of available sensor IDs

get_places

Get hierarchical map of places

get_fov_histogram

Get field-of-view occupancy over time

get_average_speeds

Get average speed metrics by direction

analyze

Perform statistical analysis on video data

Quick Start:

# Start the MCP server
nat mcp serve --config_file <path/to/config_file>

# List available tools
nat mcp client tool list --url http://localhost:9901/mcp

For complete documentation including configuration, parameters, and usage examples, see Video-Analytics-MCP Server.

On this page