🇨🇭 Part of the Swiss Public Data MCP Portfolio

🎓 global-education-mcp

MCP server for international education data – UNESCO UIS (4,000+ indicators across all member countries) and OECD Education at a Glance via SDMX. No API keys required.

🇩🇪 Deutsche Version

Overview

global-education-mcp gives AI assistants like Claude a complete international education intelligence system – literacy rates, enrolment ratios, education expenditure, teacher salaries, gender parity and SDG-4 monitoring, all accessible through a single standardised MCP interface.

The server bridges two of the most authoritative sources for internationally comparable education statistics: UNESCO UIS (global coverage, 4,000+ indicators) and the OECD's annual Education at a Glance (38 OECD countries, SDMX REST API). Both are open and require no API key.

Anchor demo query: "Compare Switzerland's education expenditure as a percentage of GDP with Finland, Singapore and South Korea over the last 10 years – and flag any SDG-4 gaps."

Features

• 🌍 UNESCO UIS – 4,000+ indicators, all UNESCO member countries, no API key

• 📊 OECD Education at a Glance – 38 OECD countries + partners via SDMX REST

• 🔍 Indicator search – browse and filter the full UNESCO indicator catalogue

• 🗺️ Multi-country comparison – benchmark any indicator across multiple countries

• 🏫 Country education profiles – 10 core indicators in one call

• 🎯 SDG-4 monitoring – structured reporting on Education for All targets

• 📈 OECD dataset search – discover and retrieve Education at a Glance dataflows

• 🔑 No API keys required – fully open data, zero setup friction

• ☁️ Dual transport – stdio for Claude Desktop, Streamable HTTP/SSE for cloud deployment

• 🛡️ Graceful degradation – API failures return helpful messages with local reference fallback

Prerequisites

• Python 3.11+

• uv (recommended) or pip

• No API keys needed

Installation

# Clone the repository
git clone https://github.com/malkreide/global-education-mcp.git
cd global-education-mcp

# Install
pip install -e ".[dev]"

Or with uvx (no permanent installation):

uvx global-education-mcp

Quickstart

# Start the server (stdio mode for Claude Desktop)
global-education-mcp

Try it immediately in Claude Desktop:

"What is Switzerland's literacy rate compared to Finland and Singapore?" "Show me education expenditure as % of GDP for CHE, DEU and AUT over the last 10 years."

Configuration

Claude Desktop Configuration

Windows (%APPDATA%\Claude\claude_desktop_config.json):

{
  "mcpServers": {
    "global-education": {
      "command": "uvx",
      "args": ["global-education-mcp"]
    }
  }
}

macOS (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "global-education": {
      "command": "uvx",
      "args": ["global-education-mcp"]
    }
  }
}

A ready-to-use claude_desktop_config.json is included in the repository root.

Cloud Deployment (SSE for browser access)

For use via claude.ai in the browser (e.g. on managed workstations without local software).

⚠️ Security note: Since v0.3, MCP_HOST defaults to 127.0.0.1. The SSE transport must always run behind a reverse proxy that adds TLS, authentication, and rate-limiting. Never expose the raw port to the internet — MCP_HOST=0.0.0.0 is only safe inside an isolated container network.

Docker (recommended):

The repository ships a hardened multi-stage Dockerfile and a docker-compose.yml that applies read_only: true, cap_drop: [ALL], security_opt: [no-new-privileges:true], and runs as non-root user uid 10001. The compose file binds the port to 127.0.0.1 so a host-level reverse proxy is required for any external access.

docker compose up --build
# then point nginx/caddy at 127.0.0.1:8000/sse with TLS + auth

Plain docker run (without compose):

docker build -t global-education-mcp .
docker run --rm \
  --read-only --cap-drop ALL --security-opt no-new-privileges \
  --tmpfs /tmp:size=16M,mode=1777 \
  -p 127.0.0.1:8000:8000 \
  global-education-mcp

Render.com:

1. Push/fork the repository to GitHub

2. On render.com: New Web Service → connect GitHub repo

3. Set environment variables in the Render dashboard:

   MCP_TRANSPORT=sse
   MCP_HOST=0.0.0.0      # Render needs 0.0.0.0; their edge layer provides TLS + auth.
   PORT=8000

4. In claude.ai under Settings → MCP Servers, add: https://your-app.onrender.com/sse

💡 "stdio for the developer laptop, sandboxed SSE container for the browser."

Available Tools

UNESCO UIS Tools

OECD Tools

Cross-Source Tools

Resources & Prompts

Resources:

• education://indicators/unesco – Quick reference for core UNESCO indicators

• education://datasets/oecd – Quick reference for OECD Education at a Glance dataflows

Prompts:

• bildungsvergleich_schweiz – Switzerland vs. Finland, Singapore, Japan

• sdg4_monitoring – SDG-4 report for CH/DE/AT

Country Codes

ISO 3166-1 Alpha-3 standard:

Example Use Cases

→ More use cases by audience →

Architecture

┌─────────────────┐     ┌──────────────────────────────┐     ┌────────────────────┐
│   Claude / AI   │────▶│   Global Education MCP       │────▶│   UNESCO UIS API   │
│   (MCP Host)    │◀────│   (MCP Server)               │◀────│   uis.unesco.org   │
└─────────────────┘     │                              │     └────────────────────┘
                        │  10 Tools · 2 Resources      │
                        │   · 2 Prompts                │     ┌────────────────────┐
                        │  Stdio | SSE                 │────▶│   OECD SDMX API    │
                        │                              │◀────│   sdmx.oecd.org    │
                        │  server.py                   │     └────────────────────┘
                        │   + api_client.py            │
                        └──────────────────────────────┘

Infrastructure Components

Caching Strategy

Project Structure

global-education-mcp/
├── src/global_education_mcp/       # Main package
│   ├── __init__.py                 # Package metadata, version
│   ├── server.py                   # FastMCP server, 10 tools, 2 resources, 2 prompts
│   └── api_client.py               # HTTP client, UNESCO UIS + OECD wrappers, formatters
├── tests/
│   ├── test_server.py              # 39 tests (basic / intermediate / advanced)
│   └── test_extended_scenarios.py  # 74 tests across 8 categories
├── claude_desktop_config.json      # Ready-to-use Claude Desktop config
├── pyproject.toml                  # Build configuration (hatchling)
├── CHANGELOG.md
├── CONTRIBUTING.md
├── LICENSE
├── README.md                       # This file (English)
└── README.de.md                    # German version

Known Limitations

• UNESCO UIS: Some indicators have sparse coverage for low-income countries or recent years

• OECD SDMX: Occasional API timeouts on large multi-country, multi-year requests; reduce the year range if needed

• OECD coverage: 38 OECD members + select partners – does not cover all UNESCO member states

• Historical depth: UNESCO UIS data availability varies by indicator; not all series go back to 1970

• Language: UNESCO UIS returns indicator labels in English only; OECD labels may vary by dataflow

• No real-time data: Both sources publish annually – figures reflect the latest published edition, not live school statistics

Compliance & Data Classification (City of Zurich)

Verbindliche Klassifikation für den Einsatz im Schulamt der Stadt Zürich (German section follows in README.de.md).

ISDS Protection Class (Stadt Zürich Schutzbedarfsklassen)

• Data owner: UNESCO UIS / OECD (external)

• System owner: Schulamt der Stadt Zürich

• Processes personal data: no

• DSG / EDÖB relevance: none (only anonymized country-level aggregates)

Schulamt Classification (BUI / Vertraulich / Streng Vertraulich)

→ The server is approved for any Schulamt use case without additional clearance from the data protection officer.

Compatibility

Major-version upgrades are deliberate decisions — the upper bounds in pyproject.toml exist so a transitive bump does not silently break the server. See CHANGELOG.md for the upgrade trail.

🛡️ Safety & Limits

Testing

# Unit tests (no API key required, no network)
PYTHONPATH=src pytest tests/ -v -m "not integration"

# Full suite including live API smoke tests
PYTHONPATH=src pytest tests/ -v

113 tests across two files and three complexity levels:

Contributing

Contributions are welcome. Please open an issue first to discuss what you would like to change.

• Follow the existing code style (Ruff linting, Black formatting)

• Add tests for new tools (tests/test_server.py or test_extended_scenarios.py)

• Use the @pytest.mark.integration marker for tests that call live APIs

• Update CHANGELOG.md and the tool table in this README

• See CONTRIBUTING.md for the full contribution guide

Changelog

See CHANGELOG.md

License

MIT License — see LICENSE

Author

Hayal Oezkan · github.com/malkreide

Credits & Related Projects

• Data: UNESCO Institute for Statistics (UIS) – open education data for all UNESCO member states

• Data: OECD Education at a Glance – annual OECD education statistics via SDMX

• Protocol: Model Context Protocol – Anthropic / Linux Foundation

• Related: swiss-transport-mcp – MCP server for Swiss public transport

• Related: zurich-opendata-mcp – MCP server for Zurich city open data

• Portfolio: Swiss Public Data MCP Portfolio