CodeGraph — VS Code Extension

Anti-hallucination context engine for AI coding agents.

CodeGraph indexes your codebase locally using embeddings and AST parsing, then exposes it as an MCP server so AI tools (Copilot, Claude, etc.) get accurate, grounded context rather than hallucinated code.

Features

Semantic Search (Ctrl+Shift+H)

Type a natural-language query and get ranked code results from your entire codebase — powered by local embeddings (no data leaves your machine).

CodeLens (GitLens-style)

See Search references and Find related code actions above every function, class, and method.

TODO / FIXME Tree

Sidebar panel scans for TODO, FIXME, HACK, BUG, NOTE, XXX comments and shows them in a priority-sorted tree.

Hover Context

Hover over any symbol to see related code snippets from other files with similarity scores.

Index Status

Live status bar showing indexing progress, chunk count, cache hit rate, and embedding metrics.

Integrations

• Jira — query issues by JQL, assignee, or status
• Slack — search messages across channels
• GitHub — query issues, PRs, and repo info
• Git log — commit history with optional file filter

Quick Start

1. Install the extension
2. Open a workspace — indexing starts automatically
3. Press Ctrl+Shift+H to search your codebase semantically
4. Click the CodeGraph icon in the activity bar to see index status and TODOs

Commands

Settings

Requirements

• VS Code 1.99+
• Node.js 18+
• codegraph-mcp server (npm install codegraph-mcp)

How It Works

1. On activation, the extension spawns the codegraph-mcp server as a child process
2. The server indexes all files using AST chunking (Tree-sitter) and generates 384-dim embeddings locally
3. Searches use hybrid Reciprocal Rank Fusion merging semantic (cosine similarity) + keyword (SQL LIKE) results
4. All data stays local — no API calls for indexing or search

License

MIT — see LICENSE