## 😃Update 08/29/2026
Our paper is publicated in **Sustainable Cities and Society** (IF=13.3)
https://www.sciencedirect.com/science/article/abs/pii/S2210670726007304

If you use this dataset or code in your research, please cite the following paper:

```bibtex
@article{WANG2026107846,
title = {LLM-empowered agents for lifeline network recovery with graph-guided MCP tools},
author = {Hongyu Wang and Shenghua Zhou and Zhengyi Chen and Wentao Wang and Lingli Li and S. Thomas Ng},
journal = {Sustainable Cities and Society},
volume = {149},
pages = {107846},
year = {2026},
doi = {https://doi.org/10.1016/j.scs.2026.107846}}
```

## 🚀🚀 Update 07/29/2026
The data and resources are updated as **highlighted🟨** below.
<img width="1042" height="727" alt="屏幕截图 2026-07-29 204727" src="https://github.com/user-attachments/assets/8b431fc0-bb05-4458-a368-a91021d8a442" />

## 🚀🚀 Update 07/29/2026
Some figures are available for public now:

### <img width="1042" height="489" alt="image" src="https://github.com/user-attachments/assets/22a28085-438c-4bea-970c-2ced3289f866" />
Fig. S5 Agents' performance on coordinating multiple contested interests for LNR
---
### <img width="1409" height="802" alt="image" src="https://github.com/user-attachments/assets/12e61bb5-138a-47b3-9e12-d79026be7ba7" />
Fig. S4 Examples of typical task failure
---
### <img width="685" height="279" alt="image" src="https://github.com/user-attachments/assets/9ecbfd8d-9001-4074-9a95-4585a24b7b73" />
Fig. S3 Comparison of outputs generated by T-ARR-01 and corresponding study 
---
### <img width="488" height="268" alt="image" src="https://github.com/user-attachments/assets/b3e1ac90-a1d2-47ff-ac1e-20d7225066f5" />
Fig. S2 Comparison of outputs generated by T-SRS-09 and corresponding study
---
### <img width="293" height="244" alt="image" src="https://github.com/user-attachments/assets/52ebf96c-2671-42d2-aa82-f21e983a6d5d" />
Fig. S1 Comparison of outputs generated by T-ICF-06 and corresponding study 
---
## 📣 Important Notice 

__⚠️ As the paper is under review, all contents in this repository are currently not permitted for reuse by anyone until this announcement is removed. Thank you for your understanding! 🙏__

## 1. Overview & Objectives

This repository contains the complete implementation, experimental data, and supplementary results for the paper **×××** developed by **XXX University** in China, and .  

Pending publication, the code is shared under a restrictive license. Once the paper is accepted, the repository will transition to a MIT license. Please contact the corresponding author for any inquiries regarding academic use during the review period.

## 2. Videos of agents operation

### 2.1 Operation of the developed prototype

↓↓↓ A demonstration of using the developed **prototype** to operate the TCG-TE LNR agents using graph-guided MCP tools

<video src="https://github.com/user-attachments/assets/62ce60a8-4f43-4ff6-a787-aa9784b2f03a" width="880"></video> 

The full video could be found here
<img width="1099" height="26" alt="屏幕截图 2026-01-03 201222" src="https://github.com/user-attachments/assets/b47b3296-cc02-40ae-acaa-9896678aeb14" />

↓↓↓ A demonstration of using the developed **prototype** to integrate a new MCP server to TCG-TE LNR agents

<video src="https://github.com/user-attachments/assets/e82f1150-2a8f-474f-b5c4-1dd9c0fcb57c" width="880"></video> 

The full video could be found here
<img width="1213" height="33" alt="屏幕截图 2026-01-03 201211" src="https://github.com/user-attachments/assets/4131b1e3-a234-4bb8-8ea5-b90479c8160b" />

### 2.2 Operation of agents based on NPG-TE pattern
#### 2.2.1 Operation of agents based on NPG-TE pattern enabled by ReAct
↓↓↓ A snippet of the operation of **NPG-TE (ReAct) agent with discrete MCP tools driven by GPT-5**.

<video src="https://github.com/user-attachments/assets/5c7c539d-9b38-4b55-abbd-5fe7da966d7c" width="880"></video> 

↓↓↓ A screenshot of Agent's response
<img width="1726" height="659" alt="image" src="https://github.com/user-attachments/assets/30e693c4-15e6-48ee-9e76-63fe79589bdf" />

The full video can be found here
<img width="1194" height="30" alt="屏幕截图 2026-01-03 201138" src="https://github.com/user-attachments/assets/6fa1f839-6f05-4dfa-97f3-2ad70f37cd24" />

↓↓↓ A snippet of the operation of **NPG-TE (ReAct) agents with discrete MCP tools driven by GPT-4o**.

<video src="https://github.com/user-attachments/assets/040dbadc-c25b-461a-9bba-7391168058cb" width="880"></video> 

↓↓↓ A screenshot of Agent's response
<img width="2000" height="601" alt="image" src="https://github.com/user-attachments/assets/825729a2-ae28-4f37-80a6-3fadba22b58d" />

The full video can be found here
<img width="1214" height="32" alt="屏幕截图 2026-01-03 201102" src="https://github.com/user-attachments/assets/aa470a0f-8f0d-4df8-a6f1-1956ad205b6f" />

#### 2.2.2 Operation of agents based on NPG-TE pattern enabled by GoT

↓↓↓ A snippet of the operation of **NPG-TE (GoT) agent with discrete MCP tools driven by GPT-5**.

<img width="1000" height="500" alt="Operation of NPG-TE (GoT)-agents driven by GPT-5" src="https://github.com/user-attachments/assets/98ef79d1-0a9c-46cf-b6fa-084dae859119" />

The full video can be found here

<img width="876" height="67" alt="屏幕截图 2026-07-30 142220" src="https://github.com/user-attachments/assets/86c73a4b-2001-4caf-839b-54225504d4cf" />

↓↓↓ A snippet of the operation of **NPG-TE (GoT) agent with discrete MCP tools driven by Deepseek-V3**.

<img width="1000" height="500" alt="Operation of NPG-TE (GoT)-agents driven by DeepseekV3" src="https://github.com/user-attachments/assets/7b461855-dd00-45d4-925b-3636d31e30e1" />

The full video can be found here

<img width="872" height="95" alt="屏幕截图 2026-07-30 142203" src="https://github.com/user-attachments/assets/7d3b3381-5917-4c31-89b8-c00fe9ba0c11" />

#### 2.2.3 Operation of agents based on NPG-TE pattern enabled by PE

↓↓↓ A snippet of the operation of **NPG-TE (PE) agent with discrete MCP tools driven by GPT-4o**.

<img width="1000" height="500" alt="Operation of NPG-TE (PE)-agents driven by GPT-4o" src="https://github.com/user-attachments/assets/1a01a8ff-8802-4f97-a3ba-23571d76bdb6" />

The full video can be found here

<img width="908" height="76" alt="屏幕截图 2026-07-30 142235" src="https://github.com/user-attachments/assets/6196d834-4d44-48b8-b1c7-fa6eb8cf1201" />

### 2.3 Operation of agents based on TCG-TE pattern

↓↓↓ A snippet of the operation of **TCG-TE agents with graph-guided MCP tools driven by Claude sonnet 3.7**.
![12月30日 (1)](https://github.com/user-attachments/assets/18879542-62bf-4bba-b764-f56af0d776d0)

The full video can be found here
<img width="1283" height="35" alt="屏幕截图 2026-01-03 201151" src="https://github.com/user-attachments/assets/51abdd29-b4fe-4004-af89-84a6294d7bda" />

↓↓↓ A snippet of the operation of **TCG-TE agents with graph-guided MCP tools driven by GPT-4.1**.

<video src="https://github.com/user-attachments/assets/8159ea48-1421-4158-b067-1bdcd8dd531e" width="880"></video> 

↓↓↓ A screenshot of Agent's response
<img width="1728" height="531" alt="image" src="https://github.com/user-attachments/assets/4359d03f-f03d-48ce-af4c-88596f86366d" />

The full video can be found here
<img width="1202" height="30" alt="屏幕截图 2026-01-03 201201" src="https://github.com/user-attachments/assets/177bccf2-53dc-4f6a-bd61-c3348f36708b" />

## 3. Repository Structure
<img width="1426" height="804" alt="image" src="https://github.com/user-attachments/assets/b15c7e1f-a918-4400-85f0-3a4e67cf1943" />


## 4. Acknowledgments

This work heavily relies on excellent open-source projects, including but not limited to:
- LangGraph & LangChain
- Hugging Face MTEB leaderboard
- NetworkX, PyTorch Geometric, and numerous LLM providers (OpenAI, Anthropic, Qwen, Llama, etc.)

We are deeply grateful to all contributors of these foundational work.