citysense 0.2.1

Geospatial RAG and MCP server for urban AI development; WUF13 aligned.

CitySense

Geospatial RAG and MCP Server Library for Urban AI Development

Specification v0.2.1 | February 2026 | WUF13 Aligned

---

Table of Contents

1. Abstract
2. Executive Summary
3. Overview
4. Architecture
5. Requirements
6. Installation
7. Quick Start
8. Configuration
9. CLI Reference
10. MCP Integration
11. Pilot Countries
12. Mathematical Foundations
13. API Summary
14. Data Sources and Connectors
15. WUF13 Alignment
16. SpatialIntent Structure
17. H3 Spatial Index
18. Dependencies
19. Data Flow
20. Troubleshooting
21. Glossary
22. Frequently Asked Questions
23. Version History
24. Development
25. Security Considerations
26. Performance Notes
27. References
28. License
29. Links

---

Abstract

CitySense is an open-source Python library that bridges geospatial urban data with large language model (LLM) toolchains. It provides two tightly integrated components: a geospatial retrieval-augmented generation (RAG) framework that semantically indexes spatial datasets and retrieves relevant context for natural language queries, and a Model Context Protocol (MCP) server that exposes city intelligence directly into AI-assisted development environments. The library is aligned with the United Nations Sustainable Development Goal 11 (Sustainable Cities and Communities), the New Urban Agenda, and the 13th World Urban Forum (WUF13) dialogue dimensions. CitySense supports five pilot countries (Azerbaijan, Finland, Sweden, Denmark, Norway) with consistent schemas and national coordinate reference systems. The framework implements spectral indices (NDVI, NDWI, NDBI, EVI, MNDWI, BSI) from Sentinel-2 L2A, Reciprocal Rank Fusion for hybrid retrieval, Urban Resilience Composite Score, and SDG 11.3.1 Land Consumption Rate. Street-level imagery (Mapillary, KartaView) and multispectral satellite data (Copernicus Data Space, Sentinel Hub) extend traditional vector geospatial analysis.

---

Executive Summary

Aspect	Description
Primary function	Semantic geospatial retrieval and MCP server for urban AI
Target users	Urban AI developers, smart city researchers, urban policy tools builders
Core technologies	Python 3.12+, Qdrant, Shapely, GeoPandas, H3, fastembed, MCP
Data sources	OpenStreetMap, Sentinel-2, Mapillary, KartaView, Copernicus Data Space
Standards alignment	WUF13, SDG 11, UN New Urban Agenda
License	EUPL-1.2

---

Overview

Design Philosophy

The core design philosophy of CitySense is that a developer should be able to express spatial intent in natural language and receive structured geospatial results without writing spatial query code. For example:

# find housing zones with low resilience scores near transit corridors in Baku

CitySense resolves such intent into structured geospatial results drawn from live and indexed sources.

Observational Layers

CitySense adds a third observational layer on top of traditional vector geospatial data:

Layer	Source	Purpose
Vector	OpenStreetMap, national registers	Buildings, roads, land use, amenities
Street-level imagery	Mapillary API v4, KartaView REST API	Ground truth, validation
Satellite	Copernicus Data Space, Sentinel Hub	Spectral indices, change detection

Key Features

Capability	Description
Natural language queries	Intent parsing and semantic retrieval over geospatial knowledge bases
Multi-source indexing	OpenStreetMap, Sentinel-2, Mapillary, KartaView, national registers
MCP integration	Native Model Context Protocol server for Cursor, Claude, VS Code
WUF13 alignment	Pilot countries with consistent schemas and national CRS
SDG 11 metrics	Land consumption rate (11.3.1), urban resilience composite score
Spectral indices	NDVI, NDWI, NDBI, EVI, MNDWI, BSI from Sentinel-2 L2A
Reciprocal Rank Fusion	Hybrid dense and sparse retrieval with configurable parameter k

Primary User Groups

Group	Use Case
Urban AI Developers	Build applications on city data with semantic access to geospatial knowledge bases
Smart City Researchers	Query, compare, and analyze datasets across multiple countries with consistent schemas
Urban Policy Tools Builders	Prototype AI-assisted planning tools aligned with UN New Urban Agenda, SDG 11, and WUF13

---

Architecture

System Stack

┌─────────────────────────────────────────────────────────────────────────────┐
│                           CitySense Stack                                    │
├─────────────────────────────────────────────────────────────────────────────┤
│  CLI (citysense)                                                             │
│  pilot init | index build | query | serve | export | imagery | watch         │
├─────────────────────────────────────────────────────────────────────────────┤
│  MCP Server (stdio/SSE)              │  Geospatial RAG Pipeline               │
│  Tools: query_spatial_context,      │  Intent Parser → H3 Filter →            │
│  analyze_housing_zone,              │  Dense Retrieval + Sparse (BM25) →       │
│  get_resilience_score, bounds       │  RRF Fusion → Reranker → Assembler      │
├─────────────────────────────────────────────────────────────────────────────┤
│  Connectors: OSM, Sentinel-2, Mapillary, KartaView, CDSE                    │
├─────────────────────────────────────────────────────────────────────────────┤
│  Vector Store (Qdrant)  │  Geometry (Shapely, GeoPandas, H3)                 │
└─────────────────────────────────────────────────────────────────────────────┘

Module Structure

Module	Responsibility
citysense.cli	Typer-based command-line interface
citysense.core	Configuration, session, logging, registry, exceptions
citysense.geo	CRS, H3, bbox, geometry, OSM utilities
citysense.connectors	OSM, Mapillary, KartaView, Sentinel, base connector
citysense.rag	Intent parsing, retriever, embedder, reranker, assembler
citysense.imagery	Street and satellite processing
citysense.urban	SDG 11 indicators
citysense.climate	Resilience scoring
citysense.pilot	Country-specific configurations
citysense.mcp	MCP server and tools

Design Principles

Principle	Implementation
Schema consistency	Pilot configs enforce national CRS and normalised GeoDataFrame schema
Extensibility	Connector and pilot registries allow plug-in modules
Observability	structlog for structured logging; configurable log levels
Type safety	mypy strict mode; Pydantic for config and data validation

---

Requirements

Requirement	Version
Python	3.12 or 3.13
Qdrant	Default endpoint http://localhost:6333
GDAL	3.10+ (for rasterio, geopandas; conda recommended)

---

Installation

pip

pip install citysense

Optional Extras

Extra	Purpose
clip	CLIP ViT-B/32 for street imagery embedding
sentinelhub	Sentinel Hub Process API
dev	ruff, mypy, pytest, mkdocs, pre-commit

pip install "citysense[clip]"
pip install "citysense[dev]"

conda-forge (Binary Compatibility)

For binary compatibility with GDAL, PROJ, GEOS:

conda env create -f environment.yml
conda activate citysense
pip install -e ".[dev]"

Editable Install

git clone https://github.com/olaflaitinen/citysense.git
cd citysense
pip install -e ".[dev]"

---

Quick Start

Minimal Workflow

citysense pilot init fi
citysense index build --city Helsinki --sources osm
citysense query "social housing zones within 1 km of metro stations in Helsinki"
citysense serve --transport stdio

With Imagery (CLIP)

pip install "citysense[clip]"
citysense pilot init fi
citysense index build --city Helsinki --sources osm,mapillary

With Sentinel Hub

pip install "citysense[sentinelhub]"
# Set SENTINELHUB_INSTANCE_ID, SENTINELHUB_CLIENT_ID, SENTINELHUB_CLIENT_SECRET
citysense index build --city Baku --sources osm,sentinel

---

Configuration

Configuration is read in priority order: environment variables (prefix CITYSENSE_), .env file at project root, defaults.

Environment Variables

Variable	Default	Description
CITYSENSE_PILOT_COUNTRY	None	Pilot module key: az, fi, se, dk, no
CITYSENSE_VECTOR_STORE_URL	http://localhost:6333	Qdrant URL
CITYSENSE_EMBEDDING_MODEL	BAAI/bge-m3	Text embedding model
MAPILLARY_ACCESS_TOKEN	None	Mapillary API token
CDSE_CLIENT_ID	None	Copernicus Data Space client ID
CDSE_CLIENT_SECRET	None	Copernicus Data Space client secret

Example .env

CITYSENSE_PILOT_COUNTRY=fi
CITYSENSE_VECTOR_STORE_URL=http://localhost:6333
MAPILLARY_ACCESS_TOKEN=your_token
CDSE_CLIENT_ID=your_id
CDSE_CLIENT_SECRET=your_secret

---

CLI Reference

Command	Description
citysense pilot init <country>	Initialize pilot configuration
citysense index build --city <name> --sources <list>	Build vector index
citysense query <natural language>	Execute semantic spatial query
citysense serve --transport stdio	Start MCP server (stdio)
citysense serve --transport sse	Start MCP server (SSE)
citysense export <format>	Export indexed data
citysense imagery <subcommand>	Street/satellite imagery operations
citysense watch	Watch mode for real-time updates

CLI Examples

# Initialize Finland pilot
citysense pilot init fi

# Build index for Helsinki from OSM only
citysense index build --city Helsinki --sources osm

# Build index with multiple sources (requires tokens)
citysense index build --city Baku --sources osm,mapillary,sentinel

# Query with natural language
citysense query "metro stations within 500m of parks in Stockholm"

# Start MCP server for Cursor/Claude
citysense serve --transport stdio

# Start MCP server with SSE for web clients
citysense serve --transport sse

---

MCP Integration

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "citysense": {
      "command": "citysense",
      "args": ["serve", "--transport", "stdio"],
      "env": {
        "CITYSENSE_PILOT_COUNTRY": "fi",
        "CITYSENSE_VECTOR_STORE_URL": "http://localhost:6333",
        "MAPILLARY_ACCESS_TOKEN": "your_token",
        "CDSE_CLIENT_ID": "your_id",
        "CDSE_CLIENT_SECRET": "your_secret"
      }
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "citysense": {
      "command": "citysense",
      "args": ["serve", "--transport", "stdio"],
      "env": {
        "CITYSENSE_PILOT_COUNTRY": "az",
        "CITYSENSE_VECTOR_STORE_URL": "http://localhost:6333"
      }
    }
  }
}

Prerequisites

1. Qdrant running at configured URL
2. Index built: citysense index build --city Helsinki --sources osm
3. API tokens for Mapillary and CDSE if using imagery

MCP Tools

Tool	Description
query_spatial_context	Execute natural language spatial query and return GeoJSON context
analyze_housing_zone	Analyze housing zone with resilience and SDG metrics
get_resilience_score	Compute URCS for a given location or zone
get_bounds	Return bounding box for pilot city or region

Transport Modes

Transport	Use case
stdio	Cursor, Claude Desktop, local scripts
sse	Web applications, remote clients

---

Pilot Countries

Country	Module Key	National CRS	Primary City
Azerbaijan	az	EPSG:32638	Baku
Finland	fi	EPSG:3067	Helsinki
Sweden	se	EPSG:3006	Stockholm
Denmark	dk	EPSG:25832	Copenhagen
Norway	no	EPSG:25833	Oslo

Pilot Configuration Attributes

Attribute	Description
country	ISO2 code
language	IETF BCP 47 tag
national_crs	EPSG string
default_cities	City name to BBox mapping
connector_priority	Ordered connector IDs
wuf13_primary_dimensions	WUF13 dimension tags
data_gaps	Known gaps and fallback strategies
informality_heuristic	SAR/spectral heuristic for tenure

---

Mathematical Foundations

Sentinel-2 Band Reference

All spectral indices are computed from Sentinel-2 L2A surface reflectance bands.

Band ID	Centre Wavelength (nm)	Resolution (m)
B02	490 (Blue)	10
B03	560 (Green)	10
B04	665 (Red)	10
B08	842 (NIR broad)	10
B11	1610 (SWIR-1)	20
B12	2190 (SWIR-2)	20

Spectral Indices

NDVI (Normalized Difference Vegetation Index)

$$\mathrm{NDVI} = \frac{\rho_{\mathrm{NIR}} - \rho_{\mathrm{Red}}}{\rho_{\mathrm{NIR}} + \rho_{\mathrm{Red}}} = \frac{B_{08} - B_{04}}{B_{08} + B_{04}}$$

Range: $[-1, +1]$. Urban green: $> 0.3$. Built-up: $< 0.1$.

NDWI (McFeeters 1996)

$$\mathrm{NDWI} = \frac{B_{03} - B_{08}}{B_{03} + B_{08}}$$

Open water: $> 0.0$.

NDBI (Zha et al. 2003)

$$\mathrm{NDBI} = \frac{B_{11} - B_{08}}{B_{11} + B_{08}}$$

Built-up: $> 0.0$.

EVI (Huete et al. 2002)

$$\mathrm{EVI} = 2.5 \cdot \frac{B_{08} - B_{04}}{B_{08} + 6 \cdot B_{04} - 7.5 \cdot B_{02} + 1}$$

MNDWI (Modified NDWI)

$$\mathrm{MNDWI} = \frac{B_{03} - B_{11}}{B_{03} + B_{11}}$$

BSI (Rikimaru et al. 2002)

$$\mathrm{BSI} = \frac{(B_{11} + B_{04}) - (B_{08} + B_{02})}{(B_{11} + B_{04}) + (B_{08} + B_{02})}$$

Reciprocal Rank Fusion

$$\mathrm{RRF}(d) = \sum_{R \in {R_{\mathrm{dense}}, R_{\mathrm{sparse}}}} \frac{1}{k + \mathrm{rank}_R(d)}$$

Default: $k = 60$.

Cross-Encoder Reranking

$$s_i = \mathrm{CrossEncoder}(q, c_i)$$

Urban Resilience Composite Score

$$\mathrm{URCS}(c) = \sum_{d \in D} w_d \cdot R_d(c)$$

where $D = {\mathrm{physical}, \mathrm{climate}, \mathrm{social}, \mathrm{infrastructure}}$.

Dimension	Weight $w_d$
Physical	0.30
Climate	0.30
Social	0.20
Infrastructure	0.20

Physical Resilience Sub-dimension

$$R_{\mathrm{physical}} = \alpha_s \cdot s_{\mathrm{condition}} + \alpha_a \cdot (1 - a_{\mathrm{norm}}) + \alpha_{\mathrm{SAR}} \cdot (1 - I_{\mathrm{informal}})$$

Default: $\alpha_s = 0.4$, $\alpha_a = 0.35$, $\alpha_{\text{SAR}} = 0.25$.

Segregation Indices

Dissimilarity Index

$$D = \frac{1}{2} \sum_{i=1}^{n} \left| \frac{g_i}{G} - \frac{m_i}{M} \right|$$

Isolation Index

$$P^* = \sum_{i=1}^{n} \left[ \frac{g_i}{G} \cdot \frac{g_i}{t_i} \right]$$

Transit Accessibility (SDG 11.2.1)

$$A(c) = \sum_{s \in S(c, r)} f(s) \cdot e^{-\lambda \cdot d(c, s)}$$

Default: $r = 800$ m, $\lambda = 0.003$.

Land Consumption Rate (SDG 11.3.1)

$$\mathrm{SDG}_{11.3.1} = \frac{\mathrm{LCR}}{\mathrm{PGR}}$$

$$\mathrm{LCR} = \frac{\ln(U_{\mathrm{land}}(t_1) / U_{\mathrm{land}}(t_0))}{t_1 - t_0}$$

$$\mathrm{PGR} = \frac{\ln(P(t_1) / P(t_0))}{t_1 - t_0}$$

Sustainable urban growth: ratio approximately 1. Ratio $> 1$: land consumption outpacing population growth. Ratio $< 1$: increasing density.

Formula Summary Table

Formula	Parameters	Default
RRF	k	60
Transit accessibility	r (radius), λ (decay)	800 m, 0.003
URCS physical	α_s, α_a, α_SAR	0.4, 0.35, 0.25
URCS dimensions	w_physical, w_climate, w_social, w_infrastructure	0.30, 0.30, 0.20, 0.20

---

API Summary

Intent Parsing

from citysense.rag.intent import parse_intent, SpatialIntent

intent: SpatialIntent = parse_intent(
    "social housing zones within 1 km of metro stations in Helsinki"
)
# intent.entity_types, intent.spatial_relations, intent.city_scope, ...

Spectral Indices

from citysense.imagery.satellite.indices import compute_index, compute_ndvi

# From band dict (B02, B03, B04, B08, B11)
ndvi_array = compute_index("NDVI", bands)
ndwi_array = compute_index("NDWI", bands)

Urban Resilience

from citysense.climate.resilience import compute_urcs, compute_physical_resilience

urcs = compute_urcs(physical=0.7, climate=0.6, social=0.5, infrastructure=0.8)
physical = compute_physical_resilience(
    street_condition=0.8, building_age_norm=0.3, informality_score=0.1
)

SDG 11 Indicators

from citysense.urban.sdg11 import compute_sdg_1131, compute_land_consumption_rate

lcr = compute_land_consumption_rate(
    urban_land_t0_km2=100, urban_land_t1_km2=120, years=10
)
ratio = compute_sdg_1131(
    urban_land_t0_km2=100, urban_land_t1_km2=120,
    pop_t0=1e6, pop_t1=1.2e6, years=10
)

Pilot Configuration

from citysense.pilot import get_pilot_config

config = get_pilot_config("fi")
# config.country, config.national_crs, config.default_cities, ...

---

Data Sources and Connectors

Connector	Data Type	API
OSM	Vector (buildings, roads, land use)	Overpass API
Sentinel-2	Multispectral imagery	Copernicus Data Space, Sentinel Hub
Mapillary	Street-level imagery	Mapillary API v4
KartaView	Street-level imagery	KartaView REST API
National registers	Cadastral, tenure	Country-specific

Connector Priority

Pilot configurations define connector_priority to resolve conflicts when multiple sources provide overlapping data.

---

WUF13 Alignment

The 13th World Urban Forum (WUF13) dialogue dimensions inform CitySense pilot configurations and indicator coverage.

WUF13 Dialogue Dimensions

Dimension	Description	CitySense Coverage
Adequate housing	Access to safe, affordable housing	Housing zone analysis, tenure heuristics
Urban planning	Land use, zoning, spatial planning	OSM land use, pilot city schemas
Climate action	Resilience, adaptation, mitigation	URCS, spectral indices, flood/heat risk
Urban economy	Employment, economic opportunity	Transit accessibility, service proximity
Urban ecology	Green space, biodiversity	NDVI, NDWI, built-up indices
Urban governance	Participation, transparency	Data standards, open schemas
Urban finance	Revenue, expenditure, investment	Indicator framework for fiscal analysis

Pilot-Specific Data Gaps

Country	Known Gaps	Fallback Strategy
Azerbaijan	Peri-urban tenure	SAR texture + NDBI heuristic for informality
Finland	Informal settlements	Not applicable; use tenure from registers
Sweden	(Per national profile)	(Per national profile)
Denmark	(Per national profile)	(Per national profile)
Norway	(Per national profile)	(Per national profile)

---

SpatialIntent Structure

The SpatialIntent dataclass captures parsed query slots from natural language input.

Attribute	Type	Description
entity_types	tuple[str, ...]	OSM-style feature types (e.g. bus_stop, railway_station)
spatial_relations	tuple[str, ...]	Relations (e.g. near, within_500m)
attributes	dict[str, str]	Attribute filters (e.g. resilience: low)
country_scope	str | None	ISO2 country code
city_scope	str | None	City name
bbox	BBox | None	Resolved bounding box
wuf13_dimension	str | None	WUF13 dialogue dimension
sdg_indicator	str | None	SDG indicator code (e.g. 11.2.1)

---

H3 Spatial Index

CitySense uses Uber H3 for hexagonal spatial indexing. Benefits include uniform cell size, hierarchical resolution, and efficient neighborhood queries.

Resolution Reference

Resolution	Avg hexagon area (km²)	Use case
5	252.9	City-scale analysis
6	36.1	District-scale
7	5.2	Neighborhood-scale
8	0.74	Block-scale
9	0.11	Building-scale

Antimeridian Handling

H3 cells crossing the antimeridian require special handling. CitySense pre-filters queries to avoid invalid geometries.

---

Dependencies

Core Dependencies

Package	Version	Purpose
shapely	>=2.1.2, <3.0	Geometry operations
geopandas	>=1.1.2, <2.0	Geospatial DataFrames
pyproj	>=3.7.1, <4.0	Coordinate transformations
pyogrio	>=0.10.0	Vector I/O
numpy	>=2.2.0, <3.0	Numerical arrays
pandas	>=2.3.0, <3.0	Tabular data
h3	>=4.1.0, <5.0	Hexagonal spatial index
rasterio	>=1.4.3, <2.0	Raster I/O
xarray	>=2023.1.0	Multidimensional arrays
pydantic	>=2.12.0, <3.0	Data validation
pydantic-settings	>=2.7.0, <3.0	Configuration
qdrant-client	>=1.17.0, <2.0	Vector store client
fastembed	>=0.4.2, <1.0	Embedding models
rank-bm25	>=0.2.2	BM25 sparse retrieval
litellm	>=1.57.0, <2.0	LLM abstraction
mcp	>=1.4.0, <2.0	Model Context Protocol
pystac-client	>=0.8.5, <1.0	STAC catalog access
rio-cogeo	>=5.4.0, <6.0	Cloud-optimized GeoTIFF
httpx	>=0.28.0, <1.0	HTTP client
Pillow	>=11.1.0, <12.0	Image processing
APScheduler	>=3.11.0, <4.0	Task scheduling
structlog	>=25.1.0, <26.0	Structured logging
typer	>=0.15.0, <1.0	CLI framework

Optional Dependencies

Extra	Packages	Purpose
clip	transformers, torch	CLIP ViT-B/32 for street imagery
sentinelhub	sentinelhub	Sentinel Hub Process API
dev	ruff, mypy, pytest, mkdocs, pre-commit	Development tooling

---

Data Flow

Index Build Flow

OSM Overpass / STAC / Connectors
        │
        ▼
Geometry + Metadata (GeoDataFrame)
        │
        ▼
H3 Cell Assignment (resolution 7 default)
        │
        ▼
Chunking + Text Serialization
        │
        ▼
Dense Embedding (fastembed) + Sparse (BM25)
        │
        ▼
Qdrant Upsert (vectors + payload)

Query Flow

Natural Language Query
        │
        ▼
parse_intent() → SpatialIntent
        │
        ▼
H3 Filter (bbox → H3 cells)
        │
        ▼
Dense + Sparse Retrieval (top-k each)
        │
        ▼
RRF Fusion (k=60)
        │
        ▼
Reranker (optional)
        │
        ▼
Assembled Context (GeoJSON, summary)

---

Troubleshooting

Issue	Cause	Resolution
Qdrant connection refused	Qdrant not running	Start Qdrant: docker run -p 6333:6333 qdrant/qdrant
Index empty after build	Connector returned no data	Check city name, pilot config, OSM coverage
Mapillary 401	Invalid or missing token	Set MAPILLARY_ACCESS_TOKEN
CDSE auth failure	Invalid credentials	Set CDSE_CLIENT_ID, CDSE_CLIENT_SECRET
CRS transform error	Mismatched EPSG	Verify pilot national_crs
H3 antimeridian error	Query crosses date line	Use bbox that does not span antimeridian

---

Glossary

Term	Definition
RAG	Retrieval-Augmented Generation; augmenting LLM context with retrieved documents
MCP	Model Context Protocol; standard for AI tools and context
H3	Uber hexagonal hierarchical spatial index
CRS	Coordinate Reference System
URCS	Urban Resilience Composite Score
LCR	Land Consumption Rate (SDG 11.3.1 numerator)
PGR	Population Growth Rate (SDG 11.3.1 denominator)
RRF	Reciprocal Rank Fusion
NDVI	Normalized Difference Vegetation Index
NDWI	Normalized Difference Water Index
NDBI	Normalized Difference Built-up Index
WUF13	13th World Urban Forum
SDG 11	UN Sustainable Development Goal 11 (Sustainable Cities)

---

Frequently Asked Questions

Is CitySense suitable for production use?

CitySense is in Alpha (Development Status 3). Use for research, prototyping, and pilot deployments. Production hardening is ongoing.

Which embedding model is used?

Default: BAAI/bge-m3. Configurable via CITYSENSE_EMBEDDING_MODEL.

Does CitySense support private repositories?

The MCP server and CLI work with any data source. Vector store (Qdrant) can be self-hosted. No data is sent to external services except configured APIs (Mapillary, CDSE, etc.).

How is coverage computed?

Coverage excludes connectors and modules under active development. Run pytest tests/unit --cov=citysense --cov-report=term-missing for current metrics.

Can I add a new pilot country?

Yes. Implement a PilotConfig in citysense.pilot, register in the pilot registry, and add to the CLI init options. See CONTRIBUTING.md.

---

Version History

Version	Date	Highlights
0.2.0	2026-02-28	WUF13 alignment, five pilot countries, SDG 11 indicators, mathematical foundations
Unreleased	-	License EUPL-1.2, initial structure per Specification v0.2.0

---

Development

Setup

git clone https://github.com/olaflaitinen/citysense.git
cd citysense
pip install -e ".[dev,clip]"
pre-commit install

Linting and Type Checking

ruff check src tests
ruff format --check src tests
mypy src/citysense

Tests

pytest tests/unit -v
pytest tests/unit -v --cov=citysense --cov-report=term-missing

Documentation

mkdocs serve

Priority Contribution Areas

Area	Description
Country connectors	Additional WUF13-relevant cities and national data sources
Non-Latin script	Arabic, Cyrillic handling for Azerbaijan and Central Asia
Informal settlement detection	Model improvements for tenure heuristics
Climate adaptation	Document parsers for national plans

Python Version Compatibility

Python	Status
3.12	Supported, tested in CI
3.13	Supported, tested in CI
3.11	Not supported (requires 3.12+)

Commit Message Format

Type	Version Bump	Example
feat:	Minor	feat(imagery): add Sentinel-3 LST connector
fix:	Patch	fix(rag): correct H3 pre-filter for antimeridian
feat!:	Major	feat!: remove deprecated search() method
docs:, chore:, test:	None	docs: add Baku pilot tutorial

---

Security Considerations

Concern	Mitigation
API tokens	Store in environment variables or secrets; never commit
Qdrant	Run locally or in private network; no default auth
Data provenance	All connectors document source and license
Dependency audit	Dependabot enabled; run pip audit periodically

---

Performance Notes

Operation	Typical scale	Notes
Index build (OSM, city)	10k–100k features	Depends on city size, H3 resolution
Query latency	100–500 ms	Dense + sparse + RRF; reranker adds ~50 ms
Embedding	BAAI/bge-m3	~50 docs/s on CPU; GPU accelerates
Qdrant	Local	Sub-10 ms for vector search at city scale

---

References

Spectral Indices

• McFeeters, S. K. (1996). The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. International Journal of Remote Sensing, 17(7), 1425-1432.
• Zha, Y., Gao, J., & Ni, S. (2003). Use of normalized difference built-up index in automatically mapping urban areas from TM imagery. International Journal of Remote Sensing, 24(3), 583-594.
• Huete, A., et al. (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment, 83(1-2), 195-213.
• Rikimaru, A., Roy, P. S., & Miyatake, S. (2002). Tropical forest cover density mapping. Tropical Ecology, 43(1), 39-47.

SDG and Urban Indicators

• UN-Habitat. (2020). SDG Indicator 11.3.1 - Land consumption rate. United Nations.
• UN-Habitat. (2020). SDG Indicator 11.2.1 - Proportion of population with convenient access to public transport. United Nations.

Retrieval

• Cormack, G. V., Clarke, C. L., & Buettcher, S. (2009). Reciprocal rank fusion outperforms condorcet and individual rank learning methods. SIGIR 2009.

Citation

If you use CitySense in academic work, please cite:

CitySense: Geospatial RAG and MCP Server for Urban AI Development.
Specification v0.2.0. WUF13 Aligned. 2026.
https://github.com/olaflaitinen/citysense

Spectral Index Interpretation Ranges

Index	Low	Medium	High	Interpretation
NDVI	< 0.1	0.1–0.3	> 0.3	Vegetation density
NDWI	< 0	0	> 0	Open water presence
NDBI	< 0	0	> 0	Built-up intensity
EVI	< 0.2	0.2–0.4	> 0.4	Enhanced vegetation
BSI	< 0	0	> 0	Bare soil / impervious

---

License

EUPL-1.2. See LICENSE.

---

Links

Resource	URL
Documentation	citysense.readthedocs.io
Repository	github.com/olaflaitinen/citysense
Issues	github.com/olaflaitinen/citysense/issues
Changelog	CHANGELOG.md
Contributing	CONTRIBUTING.md
Security	SECURITY.md

---

Acknowledgments

CitySense builds on open standards and community data: OpenStreetMap contributors, Copernicus Programme, Mapillary, KartaView, Qdrant, and the Model Context Protocol initiative. Pilot country configurations align with national spatial data infrastructures and WUF13 dialogue priorities.