Integration Guide

MCP (recommended)

AgenticVision is consumed through MCP. Any MCP client can use the same server entry.

{
  "mcpServers": {
    "agentic-vision": {
      "command": "$HOME/.local/bin/agentic-vision-mcp",
      "args": ["serve"]
    }
  }
}

After adding the entry, restart the MCP client.

With AgenticMemory

Run both servers to link visual captures to cognitive nodes:

{
  "mcpServers": {
    "agentic-memory": {
      "command": "$HOME/.local/bin/agentic-memory-mcp",
      "args": ["serve"]
    },
    "agentic-vision": {
      "command": "$HOME/.local/bin/agentic-vision-mcp",
      "args": ["serve"]
    }
  }
}

Use vision_link with a memory node ID to connect what the agent sees to what it remembers.

Agentic Flow Examples

Once the MCP server is running, your AI agent has access to visual tools. Here are example prompts and the tool chains they trigger.

Track a UI regression

Take a screenshot of the login page. Now compare it to yesterday's capture. What changed?

The agent calls vision_capture, then vision_query to find yesterday's capture, then vision_diff to compare them pixel by pixel.

Build visual evidence

Capture screenshots of every page in the checkout flow. Label them step-1 through step-5.

The agent calls vision_capture for each page with description and labels. The captures are queryable later by label or content.

Find similar UI states

I saw a layout bug last week that looks like this. Find any past captures that look similar.

The agent uses vision_similar with the current capture's embedding to find visually similar past states.

Link screenshots to decisions

Take a screenshot of this error state and link it to memory node 42 as evidence.

The agent calls vision_capture then vision_link to connect the visual evidence to a cognitive event in AgenticMemory.

Key tools available to your agent

Server runtime

Cloud servers cannot read laptop-local artifacts directly. Sync .avis/.amem/.acb to server storage first, then start MCP services there.