C — Moderate risk signals Graded October 16, 2026Several concerns worth a closer look before you install.Based on public adoption and transparency signals as of October 16, 2026. Our opinion, not a code audit or security guarantee.
AI agent skill · Web & browsing
open-science
The open-source AI research workbench for scientific research and agent workflows. Local-first, model-agnostic desktop app with extensible skills, MCP tools and connectors, Python/R execution and traceable artifacts for reproducible research on macOS, Windows and Linux.
Bottom line: Signal-based grade (3/7 points): well starred on GitHub, public source code. This reflects adoption and transparency, not a code review.
- Popularity score
- 42/100
- Rank
- #281 of 500
- GitHub stars
- 4.9k
What it does
Open-Science is a local-first, model-agnostic desktop workbench designed to support scientific research and AI agent workflows with extensible skills, MCP tools, and integrated Python/R execution. It enables researchers to build reproducible workflows with traceable artifacts, running entirely on your machine across macOS, Windows, and Linux.
Who it’s for
Research scientists, data analysts, and AI builders who need a transparent, local-first environment to prototype agent-driven workflows without cloud dependencies. Teams prioritizing reproducibility and audit trails for research outputs.
Copy-paste setup
Check these steps against the project’s own instructions before running anything.
Verify against the repo README for latest installation steps. Likely approaches:
- Desktop App: Download from GitHub Releases (opens external site) for your OS.
- Development: Clone and run locally:
bash
git clone https://github.com/aipoch/open-science.git cd open-science npm install && npm start - MCP Integration: If supporting Claude Desktop or other MCP clients, check repo for
mcpServersconfiguration examples in documentation.
Risks
- Local code execution: Python/R execution environments can run arbitrary user code; verify script sources before execution to prevent malicious code injection.
- Credential exposure: If MCP connectors or skills handle API keys/credentials, ensure they're stored securely (e.g., environment variables, OS keychain) rather than hardcoded in artifacts.
- Supply-chain trust: Dependency on extensible skills and community MCP tools; validate third-party skill origins to avoid prompt injection or data exfiltration.
- Artifact leakage: Traceable artifacts may inadvertently include sensitive research data; implement access controls and audit logging for multi-user scenarios.
- Model agnosticism complexity: Supporting multiple AI models increases surface area for model-specific prompt injection attacks; sanitize user inputs before model calls.
How to stay secure
Least-privilege permissions
- Run scripts with minimal access: When executing Python or R code, restrict what files and system resources the code can access—only grant permissions needed for that specific task.
- Isolate sensitive credentials: Store API keys and passwords in your operating system's secure storage (like Windows Credential Manager or macOS Keychain) rather than saving them in project files or notes.
- Control skill installation: Only install skills (add-on features) from trusted sources; treat community-contributed tools with caution and review what permissions they request before installing.
- Limit artifact sharing: If multiple people access the same research project, ensure only authorized team members can view or download the results files.
Network isolation
- Disconnect when not needed: Work offline or disable internet access when handling sensitive research data to prevent accidental data transmission.
- Use VPN for remote connections: If connecting to cloud services or remote AI models, route traffic through a VPN to encrypt data in transit.
- Audit external integrations: Review which MCP connectors (data link tools) have internet access, and disable those you don't actively use.
Monitoring
- Review execution logs: Check what code actually ran—before executing scripts, preview the Python or R code to confirm it matches your intention.
- Track artifact access: In shared projects, maintain a record of who accessed or downloaded research files and when.
- Monitor unusual model behavior: If the AI makes unexpected suggestions or requests unusual permissions, pause and verify the input wasn't tampered with.
Red flags to watch for
- Requests to paste credentials: Never enter API keys or passwords directly into chat—always use secure credential storage instead.
- Unexpected script suggestions: If the tool suggests running code you didn't write, carefully inspect it before execution; malicious code can steal data or compromise your computer.
- Third-party skills with vague descriptions: Avoid installing community skills that don't clearly explain what they do or where they come from.
- Results containing sensitive information: Before sharing artifact files with colleagues, scan them to ensure confidential data (personal information, passwords, unpublished findings) isn't included.
Alternatives
- Jupyter Notebook (opens external site) – established open-source notebook for reproducible research with Python/R kernels and export options.
- Observable (opens external site) – web-based reactive notebook for data visualization and collaborative research workflows.
- Nextflow (opens external site) – workflow engine for bioinformatics and scientific pipelines with strong provenance tracking.
- DVC (Data Version Control) (opens external site) – ML workflow and data versioning tool with reproducibility and artifact management.