MCP-Enabled Drug Design Agent
Develop a cutting-edge multi-agent system designed to accelerate computer-aided drug discovery or battery material design. This challenge focuses on building a sophisticated R&D workflow using graph-based agent orchestration, advanced RAG, and MCP-enabled tool integration. Agents will autonomously research, hypothesize, simulate, and refine designs by interacting with specialized scientific databases and simulation APIs. The system will employ GPT-5 for its advanced reasoning capabilities, leveraging an 'extended thinking' pattern with adaptive reasoning budgets to tackle complex scientific problems. LangGraph will define the sequential and parallel execution of agents, allowing for dynamic re-evaluation and iterative design cycles. LlamaIndex will manage an optimized RAG pipeline over vast scientific literature, ensuring agents have access to the most current and relevant data. Participants will implement MCP for seamless integration with external computational chemistry or materials simulation tools, enabling agents to execute experiments and analyze results within the defined workflow. This project simulates a real-world application of generative AI and agentic systems in scientific discovery, pushing the boundaries of autonomous research.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a graph-based multi-agent system using LangGraph with GPT-5 and LlamaIndex for autonomous scientific research and design, integrating MCP for external tools.
How work is evaluated
The evaluation will assess the agent system's ability to autonomously research, propose, simulate, and refine a scientific design (e.g., a hypothetical drug molecule or battery material) based on a given problem statement. Emphasis will be placed on the robustness of the LangGraph workflow, the efficacy of RAG, and the correct MCP tool integration.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Design and implement graph-based agent workflows using LangGraph for iterative scientific research.
Implement MCP for seamless tool integration with external scientific simulation APIs and databases.
Develop agents capable of 'extended thinking' with adaptive reasoning budgets for complex problem-solving.
Integrate an advanced RAG pipeline with LlamaIndex and GPT-5 for enhanced knowledge retrieval from scientific literature.
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
The evaluation will assess the agent system's ability to autonomously research, propose, simulate, and refine a scientific design (e.g., a hypothetical drug molecule or battery material) based on a given problem state...
Challenge input
JSON with 'problem_statement': string, 'target_properties': dict (e.g., {'stability': 'high', 'toxicity': 'low'}), 'available_scientific_papers': l...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
JSON with 'proposed_design': string (e.g., SMILES string or material composition), 'simulated_properties': dict, 'reasoning_path': list[string] (La...
- Verifies that the LangGraph workflow executed successfully without errors and produced a final output.
- Checks if the mock simulation API was invoked via MCP and its results were processed.
- Assesses if relevant information from 'available_scientific_papers' was demonstrably used in the reasoning...
- Design Validity Score target: 8
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Verifies that the LangGraph workflow executed successfully without errors and produced a final output.
- Checks if the mock simulation API was invoked via MCP and its results were processed.
- Assesses if relevant information from 'available_scientific_papers' was demonstrably used in the reasoning path or final analysis.
Metrics
- Design Validity Score · Target 8
- Reasoning Path Complexity · Target 4
Workflow tasks
- Task 1
Input: JSON with 'problem_statement': string, 'target_properties': dict (e.g., {'stability': 'high', 'toxicity': 'low'}), 'available_scientific_papers': list[string], 'mock_simulation_api_endpoint': string
Output: JSON with 'proposed_design': string (e.g., SMILES string or material composition), 'simulated_properties': dict, 'reasoning_path': list[string] (LangGraph trace), 'final_analysis': string
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.