arch7

Architecture diagram generator for Excalidraw, exposed as an MCP server.

Fork of excalidraw-architect-mcp by Bhukya Venkatesh (MIT License).

What's different

• Cylinder shapes: vertical (databases), horizontal flattened (topics/queues)
• 60+ components with auto-styling: GCP (GKE, Cloud Run, GCS, BigQuery, Pub/Sub, VPC, PSC), Kong, Confluent (Kafka, Schema Registry, ksqlDB), Dynatrace, OpenTelemetry, AI/Agentic (LLM, Agent, MCP, Vector DB, RAG)
• Smaller diamonds for agent nodes
• Repackaged as arch7_mcp

Install

pip install -e .

MCP server (Claude Code)

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "arch7": {
      "command": "python",
      "args": ["-m", "arch7_mcp"]
    }
  }
}

Tools

Tool	Description
create_diagram	Structured nodes/connections → .excalidraw
mermaid_to_excalidraw	Mermaid syntax → .excalidraw
modify_diagram	Add/remove/update nodes on existing file
get_diagram_info	Read diagram state for iteration

Usage (Python)

from arch7_mcp.core.models import Node, Edge, DiagramGraph, Direction
from arch7_mcp.engine.layout import compute_layout
from arch7_mcp.engine.renderer import build_excalidraw_file, save_excalidraw

nodes = [
    Node(id="gke", label="GKE Cluster", component_type="gke"),
    Node(id="kafka", label="Kafka Topics", component_type="kafka"),
    Node(id="pg", label="PostgreSQL", component_type="postgresql"),
]
edges = [
    Edge(from_id="gke", to_id="kafka"),
    Edge(from_id="kafka", to_id="pg"),
]

graph = DiagramGraph(nodes=nodes, edges=edges, direction=Direction.LEFT_RIGHT)
layout = compute_layout(graph)
doc = build_excalidraw_file(layout)
save_excalidraw(doc, "arch.excalidraw")

License

MIT — see LICENSE. Original copyright Bhukya Venkatesh.