OpenGraph AI

opengraph-image · static gallery

A scene memory for robot workspaces.

OpenGraph AI turns robot-camera observations into an interactive graph of objects, scenes, attributes, and spatial relationships.

Explore the example View source
scene_memory / observation_03

observation_03

├─ containsrobot_gripper

└─ holdingyellow_cloth

├─ containswork_surface

└─ on_top_oftoy_tomato

└─ in_sceneworkspace

6 nodes5 relationshipsread only

01 · Example

Robot Scene Memory

A compact visual memory built from selected BridgeData V2 observations—not a training pipeline or behavior model.

Gallery ready

Explore recurring objects across robot-camera observations.

The gallery shows a fixed 127-node graph snapshot generated offline with opengraph-image. Cytoscape.js lets you drag nodes, zoom, pan, filter, and inspect while keeping every relationship intact.

Open graph gallery 127 nodes · 323 edges
Input
Robot workspace observations
Source
BridgeData V2 · CC BY 4.0
Builder
opengraph-image
Viewer
Static HTML · Cytoscape.js
Runtime
None

BridgeData supplies realistic observations for this example. OpenGraph AI does not train on the dataset; it organizes selected images into an explorable scene memory.

02 · Install MCP

Grow a graph through conversation.

The gallery is a static snapshot. The local MCP server lets a compatible client build and query newer graph snapshots from your own images. Install uv, replace the key placeholder locally, then restart your client.

Claude Desktop

claude_desktop_config.json

{
  "mcpServers": {
    "opengraph-image": {
      "command": "uvx",
      "args": [
        "--from",
        "opengraph-image==0.1.2",
        "opengraph-image-mcp"
      ],
      "env": {
        "ANTHROPIC_API_KEY": "YOUR_ANTHROPIC_API_KEY"
      }
    }
  }
}

Windows: %APPDATA%\Claude\claude_desktop_config.json. Fully quit and reopen Claude Desktop after saving.

Cursor

.cursor/mcp.json

{
  "mcpServers": {
    "opengraph-image": {
      "command": "uvx",
      "args": [
        "--from",
        "opengraph-image==0.1.2",
        "opengraph-image-mcp"
      ],
      "env": {
        "ANTHROPIC_API_KEY": "YOUR_ANTHROPIC_API_KEY"
      }
    }
  }
}

Save as .cursor/mcp.json in a trusted project, then enable the server in Cursor’s MCP settings.

The key is required only when the MCP server extracts or queries a graph with Claude. Viewing this gallery never uses it. If the client cannot find uvx, replace it with the absolute path returned by where uvx on Windows or which uvx on macOS/Linux.

03 · Reference

Build with OpenGraph AI.

Explore the core project or install the image package directly with pip.

Python package

OpenGraph-image

pip install OpenGraph-image