feat: add Cognitive Exoskeleton MCP Server

[hanjiang-215]

Summary

Add Cognitive Exoskeleton MCP Server — a personal "second brain" MCP tool built on Hy3, featuring knowledge graph + LLM reasoning.

What it does

Extracts entities and relationships from user notes into a local SQLite knowledge graph, then leverages Hy3's reasoning capabilities to:

• Answer questions from the knowledge graph (query_mind)
• Discover hidden cross-domain connections (discover_connections)
• Detect knowledge blindspots (detect_blindspots)
• Generate cognitive portraits (analyze_cognitive_topology)
• Track concept evolution over time (trace_concept_evolution)
• Spark creative serendipity across domains (spark_serendipity)
• Ingest notes with automatic entity extraction (ingest_note)
• Recall related context during writing (recall_context)

Architecture

• MCP SDK: TypeScript SDK (@modelcontextprotocol/sdk), stdio transport
• 8 MCP tools with Zod-validated parameters and clear descriptions
• LLM: Hy3 via OpenAI-compatible API (Direct mode) or MCP Sampling protocol
• Storage: Local SQLite (sql.js, pure WASM, zero native deps)
• Privacy-first: All data stays local, API keys via environment variables only

Demo

Includes a demo video (demo-compressed.mp4) showing full workflow in CodeBuddy with Hy3.

Verification

• TypeScript type-check passes (tsc --noEmit)
• 138 tests pass across 49 test suites
• Tested on Cursor + CodeBuddy + Cline (3 MCP clients)
• CodeBuddy config examples included

Directory

All project files are under cognitive-exoskeleton-mcp/.