Curaitor Agent

AI agent for scientific data extraction

Part of Schmidt OxRSE Workshop (Sep 11–20, 2025)

---

Overview

Curaitor Agent is an AI-powered tool designed to extract, organize, and process scientific data.
It provides:

• A web interface for running the agent.
• Model Context Protocol (MCP) inspector integration to test tools and server connections.

Documentation

https://curaitor-agent-docs.readthedocs.io/latest/

---

Quick Start

Install uv

curl -LsSf https://astral.sh/uv/install.sh | sh

Clone repo

git clone git@github.com:ritesh001/curaitor-agent.git
cd curaitor-agent

Initialize project

uv sync

Edit config file

choose the model you want to use under llm:

• provider: openai
• model: "gpt-5-mini"

Provide gmail address

• send your gmail email address to nsophonrat2@gmail.com to be added to the user pool

add .env file

Create .env file in the agent folder with your

OPENAI_API_KEY=
OPENROUTER_API_KEY=
GMAIL_CREDENTIALS_PATH=
GMAIL_TOKEN_PATH=secrets/token.json

Run gmail authentication

This will work for 1 hour.

uv run python curaitor_agent_v2/gmail_create_token.py

Run web interface

uv run adk web

Run LangGraph pipeline (no Google SDK)

This runs the literature RAG workflow orchestrated by LangGraph using your config and API keys.

uv run python -m curaitor_agent.langraph_pipeline --query "your research question"

Scheduling (LangGraph)

The pipeline runs on GitHub Actions using the scheduled workflow in .github/workflows/curaitor-scheduled.yml. You can:

• Use the built-in schedule (cron) for automatic runs.
• Trigger manually via Actions → Curaitor LangGraph Weekly → Run workflow.

Outputs are uploaded to S3 under: s3://curaitor-agent-dec2025/curaitor/<run_id>/

The CLI wrapper scripts/run_daily.py runs the pipeline and upserts results into data/curaitor.sqlite by default.

Functions you can use

curaitor_agent

• create database
• query database

curaitor_agent_v2

• search and summarize paper from arxiv
• schedule time of day for daily search
• send email summary to yourself
  • send email to nsophonrat2@gmail.com to be added to the user pool

---

For Developer

Dependency Management

• Sync when requirements.txt is updated:

  uv sync

• Add a new package:

  uv add package-name

  (Don’t forget to update requirements.txt!)

---

MCP Inspector Tool

The MCP Inspector helps verify your MCP server connection and test available tools.

Requirements

• nvm (Node Version Manager)
• Node.js ≥ 18 (v22 recommended)

Setup

1. Install nvm:

   curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash
   \. "$HOME/.nvm/nvm.sh"

2. Install Node.js v22:

   nvm install 22

3. Verify versions:

   node -v   # v22.19.0
   npm -v    # 10.9.3

4. Run the MCP Inspector:

   npx @modelcontextprotocol/inspector uv run tools/mcp_server.py

5. In the MCP Inspector UI, click Connect → test tools.

---

Notes

• Ensure you’re using Node.js v22.x when running the inspector.
• Always keep your environment in sync with requirements.txt for reproducibility.

---

License

This project is licensed under the MIT License.