Indexa

Your AI reads 10,000 lines to find 1 function. Indexa gives it exactly what it needs.

A context engine for AI coding tools. Sits between your codebase and your AI assistant — returns symbols, dependencies, and execution flows instead of full files.

Index once. Query forever. 50-70% fewer tokens.

Website | npm | GitHub | VS Code Extension

Why Developers Use Indexa Daily

The problem

Every AI coding tool — Copilot, Claude, ChatGPT — has the same bottleneck: context. They open full files, skim hundreds of lines, burn thousands of tokens, and still miss dependencies across files.

Without Indexa:
  Open file (800 lines) → find 1 function → repeat x7 files
  = 10,000+ tokens burned. 30+ seconds. Missed dependencies.

The fix

Indexa indexes your codebase once, then answers any question with only the relevant symbols, their dependencies, and connections — packed within a token budget.

With Indexa:
  "explain the auth flow" → 5 symbols + deps + connections
  = 3,000 tokens. 2 seconds. Nothing missed.

Proven: 51% average token reduction on production codebases.

Quick Start

npx indexa-mcp setup

One command. Under 60 seconds. Indexes your code, configures MCP for Claude Code, runs a test query.

  Indexa Setup
  ✓ Project: my-app (typescript / react)
  ✓ Indexed 87 chunks in 5.3s
  ✓ MCP configured
  ✓ Test query: found 3 results

  Indexa ready!

Restart Claude Code. Ask anything:

• "explain the auth flow"
• "trace handleLogin"
• "what breaks if I change UserService"
• "find dead code"

What Indexa Does for Your Daily Workflow

1. Understand any feature in seconds

You:    "how does the theme system work"
Indexa: 3 symbols + connections = 1,789 tokens

  [component] ThemeSwitcher — UI picker (4 themes)
  [function]  ThemeApplier  — sets data-theme on <html>
  [export]    Theme         — type definition

  Connections:
    ThemeSwitcher → calls → useUIStore
    ThemeApplier  → calls → useUIStore

Without Indexa: Read 3 full files (400+ lines, 5,000 tokens). With Indexa: 1,789 tokens. Same understanding.

2. Trace execution flows across files

You:    "trace VendorAuthGuard"
Indexa: 9 steps across 5 files, 2 seconds

  1. VendorAuthGuard → calls VendorFlowGuard, StaffFlowGuard
  2. VendorFlowGuard → calls verifyPkceSession, setAuth
  3. verifyPkceSession → calls getAppSessionCookie, isRealJwt
  4. getAppSessionCookie → reads document.cookie
  5. useVcAuthStore → Zustand store (sessionStorage)

This normally takes 10-15 minutes of manual file-by-file tracing.

3. Know what breaks before you refactor

You:    "what breaks if I change UserService"
Indexa: 12 direct references, 23 transitive, 8 files affected

  Direct:
    [controller] UserController — src/controllers/user.ts:15
    [service] AuthService — src/services/auth.ts:42
    [component] UserProfile — src/components/UserProfile.tsx:8
    ... +9 more

  Transitive impact (depth 3):
    AuthService → LoginPage → App → 3 more files

4. Clean up your codebase

You:    "find dead code"
Indexa: 7 unreferenced symbols across 4 files

  src/utils/legacy.ts
    [function] formatOldDate (L12-28) — Unused function
    [function] parseLegacyToken (L30-45) — Unused function

  src/helpers/deprecated.ts
    [function] oldValidate (L5-18) — Unused function

You:    "find circular dependencies"
Indexa: 2 cycles detected

  Cycle 1: src/auth/guard.ts → src/auth/session.ts → src/auth/guard.ts
  Cycle 2: src/store/user.ts → src/store/ui.ts → src/store/user.ts

You:    "find duplicate code"
Indexa: 3 pairs above 92% similarity

  Pair 1 — 96.2% similar:
    [function] validateUserInput — src/validators/user.ts:12-35
    [function] validateAdminInput — src/validators/admin.ts:8-31

5. Review PRs with full context

You:    "review this PR"
Indexa: 4 changed files, 12 symbols, blast radius mapped

  src/services/auth.ts (3 symbols)
    [method] validateToken — impact: 5 direct refs, 12 files affected
    [method] refreshSession — impact: 3 direct refs, 8 files affected

  Key connections:
    validateToken → calls → decodeJWT → verifySignature
    refreshSession → calls → validateToken → updateStore

6. Export context for any AI tool

# Export to markdown — paste into ChatGPT, Claude, or any LLM
indexa export "auth flow" -o context.md

# Export as JSON — pipe to scripts, CI/CD
indexa export "payment logic" -f json | your-tool

# Copy to clipboard with structured prompt
# (VS Code: Ctrl+Shift+P → "Indexa: Copy for AI")

7. Live re-indexing — always fresh

indexa watch
# Indexa Watch Mode — monitoring ./src
# [10:30:15] 2 files → 8 chunks indexed (total: 142)
# [10:31:02] 1 files → 3 chunks indexed (total: 145)

VS Code auto-indexes on save — no manual step needed.

Efficiency Gains

How It Works

Your Code → Parser (AST) → Chunker → Embedder → .indexa/
                                                     ↓
AI Query  → Intent Router → Hybrid Search → Bundle → Response
                               ↓
                  Semantic (35%) + BM25 (25%)
                + Name match (15%) + Path (25%)

1. Index once — AST parsing (ts-morph) extracts functions, classes, components. ML embeddings (Transformers.js, 384-dim) capture meaning. Stored in .indexa/ per project.
2. Query smartly — Auto-detects intent (flow/explain/debug/search). Routes to the best strategy. No configuration needed.
3. Return minimal context — Only relevant symbols, packed within a token budget. Includes dependencies and connections between symbols.

Built for reliability

• Binary file protection — skips images, PDFs, compiled files automatically
• Corrupt index recovery — auto-backup before every save, restores from .bak
• Embedding fallback — hash-based embeddings if ML model download fails
• Buffer overflow protection — validates all byte offsets before reading
• Reverse reference index — O(1) lookups instead of O(n) scans
• 23 integration tests — first-run, recovery, binary handling, performance

18 MCP Tools

Auto-configured by indexa setup. No manual .mcp.json editing.

CLI — 21 Commands

# Setup
indexa setup                        # One-command setup
indexa doctor                       # Health check

# Search & Context
indexa search "auth middleware"      # Hybrid search
indexa bundle "payment flow"        # Context bundle for LLMs
indexa flow "handleLogin"           # Execution flow
indexa explain "pricing system"     # Code explanation

# Code Analysis
indexa dead-code                    # Unreferenced symbols
indexa blast-radius UserService     # What breaks
indexa impact-chain UserService     # Full transitive impact
indexa circular-deps                # Import cycles
indexa unused-exports               # Dead exports
indexa duplicates                   # Near-duplicate code
indexa duplicates -t 0.85           # Custom threshold

# Export & Live
indexa export "auth" -o context.md  # Export context to file
indexa export "auth" -f json        # Export as JSON
indexa watch                        # Live re-index on save

# Index Management
indexa index ./src                  # Full index
indexa update                       # Incremental via git diff
indexa clean                        # Remove junk chunks
indexa reindex ./src                # Full clean re-index
indexa serve                        # REST API on :3000

Install globally: npm i -g indexa-mcp

REST API — 21 Endpoints

Start with indexa serve:

SDK — Build on Indexa

import { createIndexa } from 'indexa-mcp/sdk';

const indexa = createIndexa({ dataDir: '.indexa' });

// Search
const results = await indexa.searchCode('auth middleware');

// Context bundle for LLMs
const bundle = await indexa.contextBundle('login flow', 3000);

// Code analysis
const dead = indexa.findDeadCode();
const cycles = indexa.findCircularDependencies();
const dupes = indexa.findDuplicates(0.90);
const impact = indexa.getFullImpactChain('UserService', 5);

// Flow tracing
const flow = await indexa.traceFlow('handleLogin', 5);

VS Code Extension

Search "Indexa Code Intelligence" in VS Code, or install from VS Code Marketplace.

Auto-starts the Indexa server. No manual terminal needed.

Why Indexa Over Alternatives

Per-Project Storage

Each project gets its own isolated index. No shared state.

my-project/
  .indexa/              <- index data (auto-gitignored)
    embeddings.json
    embeddings.json.bak <- auto-backup for recovery
    metadata.json
  .mcp.json             <- MCP config (auto-created)

Query Intent Detection

Indexa auto-classifies your query and adjusts search weights:

No configuration needed.

Documentation

Built With

• TypeScript — end to end
• ts-morph — AST parsing
• Transformers.js — local ML embeddings (all-MiniLM-L6-v2, 384-dim)
• Express — REST API
• Commander — CLI
• @modelcontextprotocol/sdk — MCP transport
• JSON storage — zero native dependencies

The Numbers

18 MCP tools
21 CLI commands
21 API endpoints
1  SDK (programmatic API)
1  VS Code extension (auto-index, AI bridge)
23 integration tests
0  API keys required
0  cloud dependencies

License

MIT — free forever, no API keys, runs 100% offline.

Built by Prashant Singh.