CodeIndexer MCP — VS Code Extension

A VS Code extension that exposes your CodeIndexer's graph search and semantic search capabilities as MCP (Model Context Protocol) tools, making them available to any LLM running in VS Code (GitHub Copilot, Gemini, etc.).

Features

10 MCP Tools

The toolset is split between cheap discovery tools (compact metadata) and targeted content tools (full source code), so an LLM can navigate a codebase without burning context tokens until it needs the actual implementation.

Tool	Returns	What it's for
`codebase_overview`	Per-repo stats, languages, semantic confidence	First call — see what's indexed
`search_code`	Compact metadata, no source	Semantic search by natural-language query
`find_symbol`	Compact metadata, no source	Find functions/classes by exact or partial name
`get_code`	Full source code	Fetch implementation of a specific element
`get_callers`	Caller list, no source	Who calls this function?
`get_callees`	Callee list, no source	What does this function call?
`get_impact`	Direct + transitive callers, affected files	"If I change X, what else might break?"
`get_call_chain`	Shortest call path A → B	How are two functions connected?
`get_file_structure`	All elements in a file with signatures	Understand a file before reading it
`find_dead_code`	Functions with zero callers	Cleanup / orphan detection

Semantic Confidence Scoring

The extension automatically computes a semantic confidence score (0–1) based on how many code elements have meaningful comments/docstrings:

Score = 0.8 → 80% of functions have docs → semantic search is highly reliable
Score = 0.2 → only 20% have docs → graph search gets 80% weight in hybrid mode

Installation

Prerequisites

The CodeIndexer Python package must be installed:
```
cd /path/to/CodeIndexer
pip install -e .
```
Node.js must be available on your system.

Install the Extension

cd vscode-extension
npm install
npm run compile
npm run package

# Install in VS Code
code --install-extension codeindexer-mcp-0.1.0.vsix

How It Works

LLM (Copilot/Gemini) → MCP Protocol → MCP Server (Node.js, stdio)
                                            ↓ HTTP
                                    Python CodeIndexer API (FastAPI)
                                            ↓
                              ┌──────────────┼──────────────┐
                          NetworkX       Milvus Lite      BM25
                         (Graph DB)    (Vector DB)     (Keyword)

Extension activates → spawns Python CodeIndexer API server
Registers MCP server definition → VS Code discovers 3 tools
LLM calls MCP tools → MCP server → Python API → search results

Configuration

Setting	Default	Description
`codeindexer.pythonPath`	Auto-detect	Path to Python executable
`codeindexer.apiPort`	`8000`	Port for the CodeIndexer API server
`codeindexer.autoStartServer`	`true`	Auto-start API server on activation

Commands

CodeIndexer: Index Current Workspace — Index the open workspace
CodeIndexer: Start Server — Manually start the API server
CodeIndexer: Stop Server — Stop the API server

Development

# Watch mode (auto-rebuild on changes)
npm run watch

# Type check
npx tsc --noEmit

# Package VSIX
npm run package

CodeIndexer MCP

Mayank Kacha