Semantic Knowledge Architect (Notion MCP)

This is a submission for the Notion MCP Challenge

What I Built

I built the Semantic Knowledge Architect, an MCP server that transforms Claude into a proactive workspace synthesizer. This agent analyzes the meaning of your notes across different Notion databases, identifies clusters of information, and maintains a "live" Mermaid.js knowledge graph. Because we all need Obsidian-style graphs. I know I do.

Features:

• Creates a Mermaid.js knowledge graph block in a specified Notion page
• Updates created knowledge graph block when new entries are added
• Creates a hardcoded 'Relation' field that shows page linkage

Video Demo

Youtube
Sample Notion Pages:

• Semantic Graph
• My Projects Database

Show us the code

Github

How I Used Notion MCP

I leveraged the Notion SDK and MCP to bridge the gap between AI reasoning and workspace structure.

Key technical implementations:

• Block-Level State Management: The agent reads page structures to find specific Block IDs, allowing it to update existing diagrams rather than just appending new ones.
• Property-Aware Querying: The query_notion_database tool was extended to detect checkbox flags, giving the user a native "remote control" for the AI's visualization logic inside Notion.
• Relational Synthesis: Claude uses its high-level reasoning to suggest and execute bi-directional links between items that share semantic similarity but live in different databases.

What Could Be Improved

• Vector Caching: Implementing a local vector store (like LanceDB) within the MCP server to cache embeddings, allowing for instant "Related Note" suggestions without re-scanning the whole database.
• Multi-Graph Support: Allowing the Architect to maintain different "views" of the same data (e.g., a Timeline Mermaid chart vs. a Flowchart) based on user tags.