🧠 OSDR Agent Workflows (MCP)

This repository demonstrates a parallel agent-based workflow using the Model Context Protocol (MCP)) to analyze NASA OSDR (Open Science Data Repository) biological datasets.

Specifically, it uses a fan-out/fan-in agent design to:

• 🛰 Fetch metadata for a given study
• 🧬 Analyze unnormalized RNA-seq counts and generate a top-10 gene plot
• 📄 Generate a clean, human-readable summary report

The agents are orchestrated in parallel using a ParallelLLM workflow and run on top of a local MCP runtime.

---

first_example/

A simple agent that uses two official MCP servers:

• mcp-server-fetch (headless browser)
• mcp-server-filesystem

osdr_example/

Custom MCP server exposing tools for interacting with OSDR data:

• osdr_fetch_metadata: Fetches metadata for a given OSDR dataset
• osdr_find_by_organism: Filters studies by organism
• Additional tools (e.g. RNA analysis) live in osdr_viz_tools

parallel_example/

The parallel agent configuration is defined in main.py and demonstrates a fan-out/fan-in model using three distinct agents:

• metadata_agent: Fetches dataset metadata via the osdr_data_fetch server.
• quant_analysis_agent: Analyzes unnormalized RNA-seq count data using osdr_viz_tools.
• summary_writer_agent: Merges both outputs and writes a Markdown summary to disk.

Uses custom MCP servers:

• osdr_data_fetch
• osdr_viz_tools

⚙️ Configuration

The cp_agent.config.yaml file controls:

• LLM backend (e.g. Ollama, OpenAI)
• MCP server connections
• Tool availability
• System prompts / metadata

🚀 Getting Started

1. Clone this repo
2. Install dependencies:

   pip install -r requirements.txt
   

3. Run an example:

   python first_example/main.py
   

📚 Resources

• Model Context Protocol)
• NASA OSDR API