Jarvis Indexer

Make all your Claude.ai conversations searchable and reviewable with AI assistance

What It Does

Captures your Claude.ai conversations, indexes them with semantic search, and lets Claude help you review what you've learned.

Simple workflow:

1. Chrome extension captures conversations daily
2. Backend stores them in PostgreSQL with vector embeddings
3. MCP server exposes data to Claude
4. You ask: "Quiz me on Python async" → Claude finds relevant conversations and creates a review session

Quick Start

Prerequisites

• Node.js 22+
• Docker & Docker Compose
• Voyage AI API key

Environment Setup

# Copy example env and fill in your values
cp .env.example .env

---

Option 1: Local Development (Recommended for Development)

Run the backend locally with just the database in Docker.

# 1. Start PostgreSQL with pgvector
docker compose up postgres -d

# 2. Install dependencies
cd backend && npm install

# 3. Run database migrations (connects to Postgres via localhost:5432)
# Only needed on first setup or after schema changes
docker compose exec app.server npm run db:migrate

# 4. Start the backend server
npm run dev

The API will be available at http://localhost:3000

---

Option 2: Full Docker Deployment

Run everything in Docker containers.

# 1. Build and start all services
docker compose up -d --build

# 2. Run database migrations (only needed on first setup or after schema changes)
docker compose exec app.server npm run db:migrate

# 3. Check logs
docker compose logs -f app.server

The API will be available at http://localhost:8080

Useful commands:

docker compose down          # Stop all services
docker compose down -v       # Stop and remove volumes (reset DB)
docker compose logs postgres # View database logs

---

Load Chrome Extension

See extension/README.md for installation instructions.

Tech Stack

Component	Technology	Why
Extension	Vanilla JS, Manifest V3	Simple, no build step
Backend	Fastify + TypeScript	Fast, async-first, type-safe
Database	PostgreSQL + pgvector	Vector search built-in
ORM	Drizzle ORM	type-safe
Embeddings	Voyage AI	High quality, $0.72/year
MCP Server	TypeScript	Exposes data to Claude
Hosting	Local → Fly.io	Free tier, zero config

Architecture

Chrome Extension → Backend API → PostgreSQL + pgvector → MCP Server → Claude

System Overview

┌─────────────────────────────────────────────────────────────────┐
│                         LOCAL MACHINE                           │
├─────────────────────────────────────────────────────────────────┤
│  ┌─────────────────┐      stdio       ┌─────────────────────┐   │
│  │  Claude Desktop │◄────────────────►│   MCP Server        │   │
│  │                 │                  │   (npx tsx ...)     │   │
│  └─────────────────┘                  └──────────┬──────────┘   │
│                                                  │              │
│  ┌─────────────────┐                             │              │
│  │ Chrome Extension│ ────────────────────────────┼──────┐       │
│  │  (claude.ai)    │                             │      │       │
│  └─────────────────┘                             │      │       │
└──────────────────────────────────────────────────│──────│───────┘
                                                   │      │
                                                   │HTTPS │HTTPS
                                                   ▼      ▼
┌─────────────────────────────────────────────────────────────────┐
│                        REMOTE SERVER                            │
├─────────────────────────────────────────────────────────────────┤
│  ┌─────────────────────────────────────────────────────────┐    │
│  │                    Backend API                          │    │
│  │  POST /api/v1/conversations     (ingest from extension) │    │
│  │  GET  /api/v1/conversations/search    (semantic search) │    │
│  │  GET  /api/v1/conversations/date-range  (date filter)   │    │ 
│  └──────────────────────────┬──────────────────────────────┘    │
│                             │                                   │
│                             ▼                                   │
│  ┌─────────────────────────────────────────────────────────┐    │
│  │              PostgreSQL + pgvector                      │    │
│  │  • conversations (metadata)                             │    │
│  │  • messages (full text)                                 │    │
│  │  • conversation_embeddings (1024-D vectors)             │    │
│  └─────────────────────────────────────────────────────────┘    │
└─────────────────────────────────────────────────────────────────┘

Data Flow

1. Capture: Chrome extension monitors claude.ai and stores conversations locally
2. Sync: Daily alarm (or manual trigger) POSTs conversations to backend API
3. Index: Backend upserts conversations and queues them for embedding generation
4. Embed: Voyage AI generates 1024-D vectors, stored in pgvector
5. Search: MCP server exposes search_conversations and get_conversations_by_date tools
6. Query: Claude Desktop calls MCP tools → API → vector similarity search → returns top 5 matches

Key Decisions

• Conversation-level embeddings (not per-message) - simpler, cheaper
• Return full conversations - Claude has 200k context window
• Single-user - no auth complexity
• Local-first option - privacy, zero cost

⚠️ Not Production Ready Authentication uses a simple Bearer token comparison—no JWT, OAuth, or OIDC. Input validation is minimal. This is a personal project; do not deploy to production or expose to untrusted users without implementing proper security measures.

Project Structure

├── backend/
│   └── src/
│       ├── api/
│       │   ├── controllers/     # Request handlers
│       │   ├── middleware/      # Auth, validation, rate-limiting
│       │   └── routes/          # Route definitions
│       ├── domain/
│       │   ├── repositories/    # DB access (Drizzle ORM)
│       │   └── services/        # Business logic
│       ├── infrastructure/
│       │   ├── db/              # Database connection & schema
│       │   └── embedding-providers/  # Voyage AI adapter
│       ├── mcp/                 # MCP server (tools, api-client)
│       ├── config/              # Environment & app config
│       ├── schemas/             # Zod validation schemas
│       ├── errors/              # Custom error classes
│       └── test/                # Unit & integration tests
│
├── extension/
│   ├── src/
│   │   ├── background/          # Service worker (sync logic)
│   │   ├── content/             # Content script (DOM capture)
│   │   ├── popup/               # Extension popup UI
│   │   ├── lib/                 # Shared utilities
│   │   └── types/               # TypeScript definitions
│   └── icons/                   # Extension icons
│
├── scripts/                     # Utility scripts
└── data/                        # Local data storage

Database Schema

conversations (id, claude_conversation_id, title, created_at, message_count)
messages (id, conversation_id, role, content, sequence_number)
conversation_embeddings (id, conversation_id, embedding vector(1024))

API Endpoints

POST /api/v1/conversations   # Extension uploads conversations
GET  /api/v1/search          # Semantic search
GET  /api/v1/conversations/:id # Get full conversation

MCP Tools

search_conversations(query, limit?)  
get_conversation(id)