Places Database

Google Maps saved places → SQLite → Claude MCP Server

Query your saved places with natural language through Claude Desktop/Code.

Quick Start

bb setup                      # Install dependencies
cp *.csv data/               # Add your Google Maps exports
bb places:import             # Import to SQLite
bb index:fts                 # Build search index
bb mcp:config                # Get Claude config

See CLAUDE.md for complete documentation.

What You Get

• SQLite database with full schema (name, coords, categories, ratings, notes)
• FTS5 search for instant text queries
• Spatial queries (find places within X km)
• MCP server for Claude integration
• Free geocoding via Nominatim/OSM
• Deduplication and smart updates

Usage

Command Line

bb query:lists               # Show all your lists
bb query:near 59.3 18.1 5   # Places within 5km
bb places:stats              # Database stats

Via Claude

Once configured:

• "Show restaurants in Stockholm"
• "Find hiking places within 10km of Abisko"
• "What places do I have in Japan?"

Architecture

CSV → import_places.py → SQLite (FTS5, spatial index)
                            ↓
                    places_mcp_server.py ← Claude Desktop/Code

Tasks

Run bb tasks to see all available commands.

Key tasks:

• places:import - Initial import
• places:update - Merge new exports
• geo:enrich - Add missing coordinates
• index:fts - Build search index
• mcp:config - Get Claude config

Files

• bb.edn - Task definitions (fully documented)
• scripts/*.py - Import, geocoding, indexing, MCP server
• data/places.db - Your SQLite database
• CLAUDE.md - Full documentation

Requirements

• Python 3.8+
• Babashka 1.3.0+
• SQLite 3.35+

Install deps: bb setup

Cost

• Import/search: $0 (local)
• Geocoding: $0 (Nominatim)
• Embeddings: ~$0.50 for 24k places (optional)

Pattern Inspiration

Adapted from Email Archive MCP

🔮 Predictions

You'll want:

• Spatial queries (works now via Haversine, faster with SpatiaLite)
• Semantic search ($0.50 for embeddings: bb cost:embeddings)
• Regular updates from new Google exports: bb places:update

The geocoding is your bottleneck—8,560 places @ 1 req/sec = 2.4 hours. Run overnight or accept current 46% coverage.

Pattern proven via your email archive. Same architecture, different domain.

---

Your maps. Your data. Your queries.