Skip to content

knowledge graph Integration

Latest

Choose a tag to compare

@BV-Venky BV-Venky released this 08 Jun 07:22

The big one: your architecture is now a living knowledge graph, not just a diagram. This release turns the project from a diagram generator into a persistent, queryable model of your system — diagrams become
views of it.

🧠 Architecture Knowledge Graph

Your services and dependencies now live in a single version-controlled markdown file (default .claude/architecture.md) that the AI builds once and reuses everywhere. It's human-readable, round-trips losslessly,
and diffs cleanly in PRs.

  • One source of truth, many views — render the full system, a single domain, or everything within N hops of one service
  • Queryable — "what breaks if payments goes down?" returns the real upstream blast radius
  • Self-checking — lint for dependency cycles, single points of failure, orphans, and dangling references
  • Bootstrap from what you have — import existing .excalidraw diagrams into the graph
  • Export anywhere — Mermaid, Graphviz DOT, or JSON
  • Git-aware — diff how the architecture changed since any ref; generate onboarding docs; detect drift against actual Python imports

🔀 Parallel edges

Two communication modes between the same services (e.g. a solid REST /charge call and a dashed Kafka payment.requested event) now render as two distinct, labelled arrows instead of collapsing into one.

🛠️ New MCP tools (kg_*)

kg_init · kg_add_service · kg_remove_service · kg_link · kg_unlink · kg_set_domain · kg_info · kg_render · kg_render_view · kg_render_around · kg_render_domain · kg_import ·
whats_connected_to · kg_path · kg_lint · kg_export · kg_diff · kg_onboarding_doc · kg_drift

🐛 Fixes

  • Edge labels no longer overlap arrows in SVG/PNG export — the line is now masked behind bound labels, matching Excalidraw's native rendering (improves every exported diagram)
  • Fixed an infinite loop in edge routing when two arrows share an endpoint

📚 Docs

  • New README section for the knowledge graph, with a kg_* tool reference
  • New architecture-knowledge-graph skill that teaches the AI how to read a codebase and build a clean graph

✅ Compatibility

Fully additive — the existing tools (create_diagram, mermaid_to_excalidraw, modify_diagram, get_diagram_info, export_diagram) are unchanged. No breaking changes.

📦 Install / upgrade

pip install -U excalidraw-architect-mcp
# or, no install:
uvx excalidraw-architect-mcp

Full changelog: https://github.com/BV-Venky/excalidraw-architect-mcp/compare/v0.3.0...v1.0.3