Vithanco for AI Assistants

Ask your assistant to map a decision, and get the diagram back as an image — not a description of one, and not a wall of Mermaid that renders into something nobody would call a Concept Map.

Vithanco ships an MCP server in two forms. Both offer the same tools; they differ only in where the work happens.

The tools

• create_concept_map — define a domain's vocabulary as falsifiable propositions.
• create_ibis — map a decision as questions, answers, pros and cons.
• create_cld — Causal Loop Diagrams, with reinforcing and balancing feedback.
• create_timeline — events over time, optionally across several tracks.
• describe_notation — hands the assistant a notation's own documentation.

What makes it more than a renderer

The assistant learns the notation. Rather than hoping a model already knows what separates a good Concept Map from a box-and-arrow sketch,

describe_notation gives it the notation's own definition: the node types, the legal edges, and the conventions that make a diagram readable.

Every diagram comes back with a critique. Alongside the image, each render returns a quality report — concepts left unconnected, a relation-to-concept ratio that suggests a thin map, propositions that do not read as sentences, edges the notation does not allow. A capable assistant reads that and improves the diagram before showing it to you.

That feedback loop is the point. You are not asking a model to draw; you are asking it to think in a notation, and giving it the means to check its work.

Local

@vithanco/vgraph-mcp runs the engine on your own machine as WebAssembly. Nothing is sent anywhere, there is no service to be down, and a diagram renders in milliseconds rather than a network round trip.

{
  "mcpServers": {
    "vgraph": { "command": "npx", "args": ["-y", "@vithanco/vgraph-mcp"] }
  }
}

Use this wherever your assistant can run a command — Claude Code, Claude Desktop, Cursor, and most editor-based assistants.

Hosted

The same tools behind a URL, for clients that cannot run a local command: Claude in the browser, ChatGPT connectors, and anything else that takes a server address and nothing more.

{
  "mcpServers": {
    "vgraph": { "url": "https://mcp.vithanco.com/mcp" }
  }
}

Which one

If you have a terminal, take the local one — it is faster, private, and depends on nothing but your own machine. If you are configuring an assistant that only accepts a URL, take the hosted one.

Both are listed in the MCP Registry as

com.vithanco/vgraph-mcp.

See also

• VGL Syntax Guide — the language the assistant writes
• Notations — what each diagram type is for
• All integrations