mcp-resilland 0.1.0

MCP server for RESILAND Intelligence — AI-powered spatial intelligence for Central Asia landscape restoration

mcp-resilland

AI-powered spatial intelligence for landscape restoration — now in any MCP client.

Built for the World Bank's $256M RESILAND CA+ program covering Uzbekistan, Kazakhstan, Kyrgyzstan, Tajikistan, and Turkmenistan.

---

mcp-resilland is a Model Context Protocol server that makes the full power of the RESILAND Intelligence platform available to Claude Desktop, Cursor, Continue, Windsurf, Zed, and any other MCP-capable client.

Ask your AI assistant questions like:

"Analyze cadastral parcel 1248 in Samarqand for drought-tolerant restoration, cite the Phase 4 policy documents, and generate a feasibility report in Turkish."

…and the model will call the right sequence of tools — pulling cadastral geometry, 12-month Sentinel-2 NDVI, OSM infrastructure proximity, country climate indicators, curated species fit, and pgvector-backed policy RAG — then synthesize the answer with Opus 4.7 on the RESILAND backend.

---

Features

• 🗺️ 3,563 cadastral forest parcels (Uzbekistan pilot) + all 5 RESILAND CA+ countries
• 🌱 Curated species catalog — 64 Central Asia tree/shrub species with climate profile, tolerances, restoration priorities, nursery availability
• 🛰️ Sentinel-2 NDVI history — 12-month median per parcel from local Cloud-Optimized GeoTIFFs
• 🌡️ Country climate indicators — ERA5 temperature, CHIRPS precipitation, GWIS FWI fire danger, SPI-3 drought, WRI water stress, Hansen tree-cover loss, MODIS dust
• 🛣️ OSM infrastructure proximity — nearest road / settlement / water body across ~370K features for all 5 countries
• 📄 Phase 4 policy RAG — pgvector semantic search across World Bank Phase 4 feasibility and nursery handbooks
• 🤖 Opus 4.7 feasibility reports — 10-section Phase 4-style markdown with citations and token/cost accounting
• 🌐 Four-language output — English · Türkçe · Русский · Oʻzbekcha

---

Install

pip install mcp-resilland

Or, if your MCP client supports it (recommended for Claude Desktop):

uvx mcp-resilland

---

Configure

Create an API key on the RESILAND dashboard: https://resilland.com/settings/api-keys

(Keys are free with soft daily/weekly/monthly quotas — no credit card.)

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "resilland": {
      "command": "uvx",
      "args": ["mcp-resilland"],
      "env": {
        "RESILLAND_API_KEY": "rsi_live_..."
      }
    }
  }
}

Cursor / Continue / Windsurf / Zed

Any MCP-capable client. Point the stdio transport at the mcp-resilland binary with the same environment variable.

Self-hosted backend

If you run your own RESILAND Intelligence instance, override the upstream:

export RESILLAND_BASE_URL=http://localhost:8000
export RESILLAND_API_KEY=rsi_live_...
mcp-resilland

---

Capabilities

Tools (9)

Tool	What it does
get_balance	Current API key quota state + totals
list_models	Available models + supported locales/countries
list_parcels	Cadastral parcel search (country / viloyat / tuman / bbox)
get_parcel	Parcel detail + nearest OSM + nearest nurseries
list_nurseries	Nurseries across the RESILAND region
get_nursery	Single nursery detail
analyze_area	Full spatial context: NDVI · climate · species · infrastructure · RAG
generate_feasibility_report	Opus 4.7 Phase 4-style markdown report with citations
translate_brief	EN ↔ TR ↔ RU ↔ UZ translation, markdown-preserving

Resources (4)

Resources are URI-addressable, read-only views the host can surface as "attach to chat" items, sidebar references, or autocomplete targets.

URI	Returns
resiland://info	Server version + runtime config
resiland://parcels/{parcel_id}	Parcel detail (same shape as get_parcel)
resiland://nurseries	Full nursery collection
resiland://species/{species_id}	Curated species record (climate profile, tolerances, restoration use-cases)
resiland://climate/{country}	Country-level climate indicators

Prompts (2)

Pre-canned conversation starters the user can pick from their client's prompt picker.

Prompt	Purpose
restoration_brief	Produce a structured restoration recommendation for an area
species_matchmaker	Pick tree species matching given site conditions

---

Example conversations

You:   List all RESILAND-9 nurseries in Kazakhstan.
Claude: [calls list_nurseries(country="KAZ", category="RESILAND-9")]
        Here are the 9 RESILAND program nurseries in Kazakhstan…

You:   For parcel 1248, what are the top three drought-tolerant species?
Claude: [calls analyze_area(parcel_id=1248, query="drought tolerant")]
        Based on the curated species fit and Phase 4 RAG citations,
        the top three are Haloxylon ammodendron, Elaeagnus angustifolia,
        and Ulmus pumila…

You:   Generate a full feasibility report for the Dargom nursery in Turkish.
Claude: [calls generate_feasibility_report(nursery_id="…", locale="tr")]
        I've prepared a 2,300-word Turkish feasibility report.
        Estimated cost: $0.08. Sources cited: 7 Phase 4 chunks…

---

Architecture

    MCP Client (Claude Desktop / Cursor / Continue / Windsurf / Zed)
                          │  stdio (JSON-RPC)
                          ▼
                    mcp-resilland
              (FastMCP 3.x · Python 3.10+)
                          │  httpx async
                          ▼
              RESILAND Intelligence backend
                  https://resilland.com/v1/public/*
                          │
          ┌───────────────┼────────────────┐
          ▼               ▼                ▼
       PostGIS         pgvector         Opus 4.7
   (parcels · OSM)   (Phase 4 RAG)    Sonnet 4.6
                                       Haiku 4.5

The MCP server is a stateless proxy. It does not compute, store, or cache — all state lives on the RESILAND backend (PostGIS for geometry, pgvector for RAG, MinIO for tiles, Redis for the Anthropic response cache).

---

Development

git clone https://github.com/ilhankilic/mcp-resilland
cd mcp-resilland
pip install -e ".[dev]"

# Run against a local RESILAND backend:
RESILLAND_BASE_URL=http://localhost:8000 RESILLAND_API_KEY=rsi_live_... \
    python -m mcp_resilland

# Run tests:
pytest

Project layout

mcp-resilland/
├── pyproject.toml           # hatchling · Apache-2.0 · Python 3.10+
├── README.md                # this file
├── LICENSE                  # Apache 2.0
├── src/mcp_resilland/
│   ├── __init__.py          # package metadata
│   ├── __main__.py          # mcp-resilland console script entrypoint
│   ├── server.py            # FastMCP instance + module wiring
│   ├── client.py            # httpx async client (bearer auth)
│   ├── tools.py             # 9 MCP tools
│   ├── resources.py         # 4 MCP resources (URIs)
│   └── prompts.py           # 2 MCP prompts
└── tests/                   # pytest + respx

---

Authors

• İlhan Kılıç — lead engineer, architect
• Enis Kaan Keskin — spatial data, biodiversity modeling

Acknowledgments

We are grateful for the invaluable advisory support of:

• İsmail Belen — President, United Nations Forum on Forests (UNFF)
• Doç. Dr. Hülya Hernandez — FAO Türkiye Office (Food and Agriculture Organization of the United Nations, Türkiye Office)

Their guidance on international forest policy frameworks, landscape restoration standards, and Central Asia ecological context was instrumental in shaping this work.

We also thank the World Bank for the RESILAND CA+ Phase 4 feasibility documentation that forms the semantic-search corpus of this platform.

---

Links

• Dashboard: resilland.com
• PyPI: pypi.org/project/mcp-resilland
• Issues: github.com/ilhankilic/mcp-resilland/issues
• Model Context Protocol: modelcontextprotocol.io
• FastMCP: gofastmcp.com
• World Bank RESILAND CA+ Program: worldbank.org

---

License

Licensed under the Apache License, Version 2.0. See LICENSE.

Copyright © 2026 İlhan Kılıç and Enis Kaan Keskin.