Content
# GEP-AI
**GEP-AI Environmental Education Lesson Design MCP Server** — Design environmental education lessons with your existing AI (Claude, Gemini, ChatGPT/Codex) without subscribing to a separate service.
Developed as part of a research project by the Ministry of Education and the Korea Environmental Preservation Association, the GEP-AI platform ([What is GEP-AI?](#what-is-gep-ai)) has been repackaged as an MCP (Model Context Protocol) server.
## What is GEP-AI?
GEP-AI is an AI-based environmental education lesson design platform designed and developed by the National Gongju University Industry-Academic Cooperation Group in a research project commissioned by the Ministry of Education and the Korea Environmental Preservation Association (_Greening Education Partnership with AI_).
### Problem Awareness — "It's not that there aren't many materials, but they don't fit my class."
The research team analyzed environmental education materials from various educational offices and found that the format, structure, and classification system varied greatly, making it difficult to connect and utilize materials. Most materials were in non-structured formats like PDF and HWP, making search and combination challenging.
### Design Philosophy — "Structure is provided by AI, and meaning and judgment are made by teachers."
- **Support for thinking, not replacement**: AI does not automatically generate answer-type lesson plans. Instead, it provides a draft that suggests direction and options based on the teacher's conditions and context, with the final judgment and educational responsibility always lying with the teacher (Human-in-the-Loop).
- **From search to semantic connection**: The platform aims for a 'rhizome-type DB' where achievement standards, themes, activities, and local resources are connected in a network, allowing teachers to combine and generate lessons in their own context.
## Tool List
## GEP-AI
### Problem Awareness —
### Design Philosophy —
### Research Prototype → This MCP Server
The research prototype was implemented as a web platform with a HITL multi-agent workflow, 9-dimensional ontology-based RAG hybrid search, and 5-item quality verification. However, due to the high operational cost of calling its own LLM, the core assets were repackaged into this MCP server.
| Research Prototype (Web Platform) | This MCP Server |
|---|---|
| Self-LLM (Gemini) call → Operational cost occurs | **User's AI** (Claude·ChatGPT·Gemini) handles LLM → Operational cost 0 |
| HITL multi-agent forces query and review stages | 6 prompts enforce the same principle ("one question at a time, summary confirmation before generation") |
## What's Included
| Asset | Scale | Description |
|---|---|---|
| Achievement Standards DB | 3,285 items | 2022 revised curriculum achievement standards for all subjects in elementary, middle, and high school |
| Environmental Education Resource Catalog | 933 items | Materials from 12 educational offices across the country (education department resource map) — mapping by school level, subject, environmental theme, SDGs, and achievement standards |
| **Original Document Catalog + Full Text** | 1,113 items / 22,341 chunks | Text of educational office materials in PDF format — full-text search (BM25) and document reading support |
| Learning Wheel Model | — | Professor Jae-young Lee's environmental education learning model (4 stages + 4 connection rings) |
| Lesson Design Procedure | — | 9-step design prompts verified in GEP-AI actual service |
All data is embedded in the package, **no external DB or API key required**.
### Embedded DB Details
**① Achievement Standards DB** (`standards.json`, 3,285 items)
- Source: 2022 revised curriculum (Ministry of Education notification)
- Composition: 653 elementary, 580 middle school, 2,052 high school, 160 subjects
- Fields: `code`, `school`, `subject`, `grade`, `content`
**② Environmental Education Resource Catalog** (`resources.json`, 933 items)
- Source: Education department resource map input data — 12 educational offices
- Fields: `office`, `collection`, `title`, `materialType`, `activityTypes`, `schoolLevels`, `hours`, `subjects`, `competencies`, `place`, `envTopics`, `sdgs`, `methods`, `standards[{code, text}]`
**③ Original Document Catalog + Full Text** (`documents.json` 1,113 items + `chunks.jsonl.br` 22,341 chunks)
- Source: GEP-AI actual service RAG pipeline — 1,113 educational office material PDFs with AI metadata enhancement
- Fields: `fileName`, `region`, `schoolLevel`, `subjects`, `envTopics`, `sdgs`, `competencies`, `methods`, `resourceType`, `activityType`, `keywords`, `standards`, `resourceIds`, `fileUrl`
## Tools
| Tool | Description |
|---|---|
| `search_standards` | Achievement standard search (keyword, school level, subject, grade, code) |
| `search_resources` | Environmental education resource search (keyword, school level, subject, environmental theme, SDGs, educational office) |
| `get_resource_detail` | Resource details + linked achievement standards + original document guide |
| `map_standard_to_resources` | Achievement standard code → linked resource list |
| `search_fulltext` | **Full-text search of resource original documents** — search activity methods, experiment procedures, and cases in PDF text (BM25) |
| `get_document_text` | Read original document — read the text of the searched document in chunks |
| `explore_topics` | Explore environmental theme maps — number of resources by theme, related themes, SDGs, achievement standards, and keywords |
| `related_resources` | Recommended similar resources — provide shared achievement standards, themes, and subjects |
## Prompts
| Prompt | Description |
|---|---|
| `design_lesson` | Entire procedure for lesson design based on the learning wheel (information collection → achievement standards → resources → lesson plan → evaluation → activity sheet) |
| `context_interview` | **Preventive interview** — one question at a time about lesson intentions, student context, classroom/school/community context, evaluation evidence, and expected misconceptions |
| `local_pbl_coach` | Design project-based learning (PBL) based on actual environmental problems in schools and local communities — driving question, student role, real audience, and process evaluation |
| `reflect_lesson` | Evidence-based reflection after lesson execution — diagnose causes with refutable hypotheses, and minimal modifications for the next class (next cycle of the learning wheel) |
| `lesson_brief` | Summarize design conversations into a one-page brief (with undetermined marks, for colleague sharing and handing over to the next AI task) |
| `learning_wheel_guide` | Explanation of the learning wheel model |
## Installation
> 🧑🏫 **Getting started?** Follow the [easy installation guide (docs/INSTALL.md)](docs/INSTALL.md) for a step-by-step guide. It includes installation without Node.js, detailed procedures by platform, and troubleshooting.
Local installation requires Node.js 18 or higher.
### Claude Code
```bash
claude mcp add gepai -- npx -y gepai-mcp
```
### Claude Desktop
**Settings → Developer → Edit Config** to open `claude_desktop_config.json` (Windows: `%APPDATA%\Claude\`, macOS: `~/Library/Application Support/Claude/`) and add, then restart the app:
```json
{
"mcpServers": {
"gepai": {
"command": "npx",
"args": ["-y", "gepai-mcp"]
}
}
}
```
### Gemini CLI
Add to `~/.gemini/settings.json`:
```json
{
"mcpServers": {
"gepai": {
"command": "npx",
"args": ["-y", "gepai-mcp"]
}
}
}
```
### Codex (CLI · IDE extension · Desktop app)
One line in the terminal:
```bash
codex mcp add gepai -- npx -y gepai-mcp
```
Or add to `~/.codex/config.toml` (IDE extension: ⚙ menu → **MCP settings → Open config.toml**). CLI · IDE extension · Desktop app (Windows · macOS) all share this configuration file:
```toml
[mcp_servers.gepai]
command = "npx"
args = ["-y", "gepai-mcp"]
```
Verification: Run `codex` and then `/mcp`.
### Cursor
Add to global `~/.cursor/mcp.json` or project `.cursor/mcp.json`:
```json
{
"mcpServers": {
"gepai": {
"command": "npx",
"args": ["-y", "gepai-mcp"]
}
}
}
```
### Google Antigravity (CLI · IDE)
Antigravity product family shares one integrated MCP configuration file between CLI and IDE: `~/.gemini/config/mcp_config.json`
```json
{
"mcpServers": {
"gepai": {
"command": "npx",
"args": ["-y", "gepai-mcp"]
}
}
}
```
- **IDE**: Agent side panel top **"… → MCP Servers → Manage MCP Servers → View raw config** to edit the same file. (Antigravity 2.0: **Settings → Customizations → Installed MCP Servers → Add MCP**)
- To use per project, place `.agents/mcp_config.json` in the repository root. You can check server status and connection logs with `/mcp` in the prompt.
## Usage Examples
After installation, try saying to your AI:
- "Design a 4th-grade climate change lesson plan" (or run `design_lesson` prompt)
- "Find achievement standards related to ecosystems in middle school science"
- "Are there educational office materials available on the topic of plastic?"
- "[6과05-03] Show me resources linked to this achievement standard"
## Remote MCP — Register URL without Installation (Recommended for Teachers)
If Node.js installation is difficult, register the public server URL:
**`https://gepai-mcp.vercel.app`**
| Platform | Registration Method |
|---|---|
| Claude (Web · Desktop) | Settings → Connector → **Add custom connector** → Enter the URL above |
| ChatGPT | Settings → Connector (Developer mode) → New connector → Enter the URL above |
| Codex (CLI · IDE · Desktop app) | `[mcp_servers.gepai]` + `url = "https://gepai-mcp.vercel.app"` in `~/.codex/config.toml` — All Codex products share this file |
| Gemini CLI | Add to `~/.gemini/settings.json` under mcpServers: `"gepai": { "httpUrl": "https://gepai-mcp.vercel.app" }` |
| Google Antigravity | Add to `~/.gemini/config/mcp_config.json`: `"gepai": { "serverUrl": "https://gepai-mcp.vercel.app" }` — Note key names: `url` · `httpUrl` not supported |
It operates on a free tier, performing only searches and not calling LLMs.
### Direct Hosting
```bash
npx gepai-mcp --http --port 3737
# Health check: GET /healthz, MCP endpoint: POST /
# If PORT environment variable is present (e.g., Vercel), automatically starts in HTTP mode.
```
## Development
```bash
npm install # Install dependencies (auto-builds in prepare)
npm run build:data # Re-generate data/source/*.csv → data/generated/*.json
npm run build # TypeScript compilation
npm test # Smoke test (MCP client-server roundtrip)
```
```
data/source/ Original CSV (Achievement standards, Resource map catalog)
data/generated/ Built JSON (bundled in package, committed)
src/ MCP server (tools, prompts, search)
src/prompts/ Lesson design procedure prompts
```
## Data Source and License
- **GEP-AI Research**: Final report of the research project "Development of Environmental Education Resource Map and AI-based Teacher Lesson Design Utilization System" (commissioned by the Ministry of Education and Korea Environmental Preservation Association, conducted by Chungnam National University Industry-Academic Cooperation Foundation, lead researcher Jae-young Lee, 2026. 2)
- **Code**: MIT License
- **Achievement Standards**: 2022 revised curriculum (public data announced by the Ministry of Education)
- **Resource Catalog**: Metadata based on input data from the Ministry of Education's resource map. The original documents are copyrighted by each provincial office of education, and this server only provides metadata (title, source, classification).
- **Learning Wheel Model**: Prof. Jae-young Lee's learning model. This server only includes a summary of the model's structure.
## Roadmap
- [x] npm package distribution (`npx gepai-mcp`) — [gepai-mcp@0.1.0](https://www.npmjs.com/package/gepai-mcp)
- [x] Remote MCP deployment — https://gepai-mcp.vercel.app (Vercel, teachers can use by registering URL)
- [x] In-depth search of original documents — v0.3.0: built-in BM25 expert search with 22,341 chunks of original documents (no embedding needed, semantic re-ranking performed by calling AI)
- [x] Initial securing of original document links — v0.4.0: 41 official page URLs of educational resources (educational office resources, webzines, e-books, verified by web) → connected to 517 resources and 487 documents
- [x] Direct links to original PDF documents — v0.5.0: posting educational office PDFs to Google Drive storage and linking file URLs to 997/1,053 documents (95%)
- [x] Installation guides for various platforms — [docs/INSTALL.md](docs/INSTALL.md) (verified with Claude, ChatGPT, Codex, Gemini, Antigravity, Cursor, and official documentation)
- [ ] Enhancing installation guide screenshots
*GEP-AI: Results of the Ministry of Education's environmental education research project, made available for everyone.*
MCP Config
Below is the configuration for this MCP Server. You can copy it directly to Cursor or other MCP clients.
mcp.json
Connection Info
You Might Also Like
No recommendations available