HugeContext

Dynamic, Intent-Aware Semantic Memory for LLMs

HugeContext is a VS Code extension and standalone MCP server that acts as a local Context Engine for AI assistants. It provides intelligent, contextual code understanding for Claude Desktop, Cursor, and other MCP-compatible tools.

Unlike cloud-based solutions, HugeContext runs entirely locally - your code never leaves your machine.

After Installing in VS Code (2-minute setup)

1. Install the VSIX (hugecontext-0.1.0.vsix) via VS Code or code --install-extension hugecontext-0.1.0.vsix.
2. Open your project folder in VS Code.
3. First launch downloads the embeddings model (~150 MB); keep internet on. Watch progress in Output → HugeContext.
4. Wait for the initial index (starts ~5s after opening). Look for “Indexing complete” in Output → HugeContext.
5. Run Cmd/Ctrl+Shift+P and use:
   • HugeContext: Query Context for semantic search.
   • HugeContext: Show Context Graph to browse the graph.
   • HugeContext: Show Indexing Stats to verify readiness.
   • HugeContext: Reindex Workspace after big changes.
   • HugeContext: Start MCP Server to connect Claude Desktop/Cursor or any MCP client.
6. If things seem stuck, open Output → HugeContext and rerun Reindex Workspace.

Pick your workflow (beginners and power users)

• Guided panels (visible UI)

  • Keep Output → HugeContext open to see status and query results.
  • Open HugeContext: Show Context Graph for visual navigation; refreshes every ~5s.
  • Use HugeContext: Show Indexing Stats to confirm the index is ready before querying.

• Invisible mode (hotkeys, zero panels)

  • Add keyboard shortcuts in Preferences: Keyboard Shortcuts for:
    • hugeContext.queryContext → your main hotkey (e.g., Cmd/Ctrl+Shift+L).
    • hugeContext.startMcpServer for Claude/Cursor integration.
    • hugeContext.reindex to refresh without menus.
  • Run queries and only open Output → HugeContext when you need to view results.

• MCP/CLI mode (power users)

  • Run node dist/mcp-cli.js --workspace /path/to/project and register it in your MCP client.
  • Works even without opening VS Code; acts as a local context engine.

Two Ways to Use

1. VS Code Extension

Install in VS Code for real-time code indexing with heat tracking, graph visualization, and integrated queries.

2. Standalone MCP Server

Run mcp-cli.js directly with Claude Desktop or any MCP client - no VS Code required.

Features

Intent-Aware Context Retrieval

Automatically detects query intent and adjusts context retrieval strategy:

• Bug/Debug: Prioritizes error-related code, stack traces, recent changes
• Refactor: Focuses on related files, dependencies, and architectural patterns
• Understand: Returns comprehensive documentation and code structure
• Tests: Surfaces test files and test utilities
• Architecture: Provides high-level system overview

Semantic Code Search

• Uses all-MiniLM-L6-v2 embeddings for semantic similarity
• Finds conceptually related code even with different terminology
• Supports 9 programming languages

Context Hypergraph

Builds rich relationships between code entities:

• File → Function/Class → Method relationships
• Import/export dependencies
• Test ↔ Implementation connections
• Commit → File change relationships

Dynamic Heatmap

Tracks developer activity in real-time:

• Recently viewed/edited files get higher "heat"
• Heat propagates to related code through graph edges
• Prioritizes contextually relevant code in results

Git-Aware RAG

Uses Git history for episodic memory:

• Indexes commit messages and diffs
• Detects files that frequently change together
• Surfaces relevant historical context

MCP Server

Exposes tools for AI assistants:

• query_context - Semantic search with intent detection
• hugecontext_status - Index statistics and health check
• Compatible with Claude Desktop, Cursor, and other MCP clients
• Returns ranked, formatted context chunks

Incremental Indexing

Efficient file watching system:

• Debounced updates (groups rapid changes)
• Content-hash based change detection (only re-indexes if content changed)
• Automatic cleanup on file deletion

Quick Start

# Clone the repository
git clone https://github.com/yourusername/HugeContext.git
cd HugeContext

# Install dependencies
npm install

# Download Tree-sitter WASM parsers
npm run download-wasm

# Build the extension
npm run build

# Run in VS Code (F5)

Configuration

VS Code Settings

MCP Configuration (Claude Desktop)

Add this to your ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "hugecontext": {
      "command": "node",
      "args": [
        "/path/to/HugeContext/dist/mcp-cli.js",
        "--workspace",
        "/path/to/your/project"
      ]
    }
  }
}

Example for a specific project:

{
  "mcpServers": {
    "hugecontext-myapp": {
      "command": "node",
      "args": [
        "/Users/yourname/HugeContext/dist/mcp-cli.js",
        "--workspace",
        "/Users/yourname/projects/my-app"
      ]
    }
  }
}

Available MCP Tools:

• query_context - Semantic search with intent detection
• hugecontext_status - Get indexing statistics

Supported Languages

Commands

Performance

Contributing

Contributions welcome! Guidelines coming soon.

License

HugeContext is provided under the HugeContext Internal Use License 1.0.0 (see LICENSE for full terms).

Acknowledgments

• Tree-sitter - Incremental parsing
• Transformers.js - ML embeddings
• sql.js - In-browser SQLite
• Model Context Protocol - AI integration standard