</>MCP Agents Market
Agent

openscience

by synthetic-sciences3.4kTypeScriptUpdated 2026-09-01

The open-source AI workbench for scientific research

OpenScience is an autonomous AI workbench designed for conducting scientific research across machine learning, biology, physics, and chemistry. When given a research goal, it independently performs literature reviews, forms hypotheses, writes and executes code, runs experiments, queries scientific databases, and produces written analyses. The system includes 309 bundled skills for training models, molecular biology, cheminformatics, and scientific computing, plus 42 database connectors for resources like UniProt, PubChem, arXiv, and Semantic Scholar. It runs locally in your browser and supports frontier models via API keys, local models, or optional Ace-managed routing.

Key Features

End-to-end research automation: literature review, hypothesis formation, code execution, experimentation, analysis, and write-up in one continuous session
309 bundled scientific skills covering DeepSpeed, PEFT, TRL training, molecular biology, cheminformatics, papers, LaTeX, and scientific runtimes
42 built-in scientific database connectors including UniProt, PDB, Ensembl, ChEMBL, PubChem, arXiv, OpenAlex, and Semantic Scholar
Browser-based workspace with file tree, editor, terminal, session history, and inline rendering for molecules, structures, genomes, and plots
Adaptive delegation modes (Off, Auto, High) for parallel research tasks and exploration without fixed worker quotas
Model-agnostic architecture supporting OpenAI, Anthropic, Gemini, DeepSeek, local models, or optional Ace credit-backed routing
Extensible via LSP integration, MCP servers, plugins, custom agents, commands, and TypeScript SDK
Sandboxed execution using macOS Seatbelt or Linux bubblewrap for code containment with configurable approval levels

Use Cases

  • 01Conducting machine learning experiments with automated training, evaluation, and hyperparameter optimization
  • 02Performing bioinformatics research by querying protein databases, analyzing genomic sequences, and running molecular simulations
  • 03Accelerating chemistry research through automated cheminformatics queries and molecular structure analysis
  • 04Conducting literature reviews across arXiv, Semantic Scholar, and OpenAlex with automated hypothesis generation
  • 05Running reproducible physics simulations with automated code generation, execution, and result visualization
  • 06Writing scientific papers with automated LaTeX generation, figure creation, and reference management

Related Agents

View more

openscience — FAQ

What is OpenScience?+

OpenScience is an autonomous AI agent for scientific research that performs the complete research loop—from literature review through experimentation to write-up—when given a research goal. It includes 309 scientific skills and 42 database connectors, runs locally in your browser, and works with frontier or open-weight models.

How do I install OpenScience?+

Install globally with npm using 'npm install -g @synsci/openscience' then run 'openscience' to launch the workspace, or use 'npx synsci' for a single-command launch. Platform-specific desktop binaries are also available from GitHub Releases for macOS, Linux (kernel 5.1+), and other platforms.

What API keys or prerequisites are required?+

You need at least one model provider API key (ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, or DEEPSEEK_API_KEY) or access to local models. A free Synthetic Sciences account is required to link the installation and maintain session history. Optional Ace provides pay-as-you-go credit-backed models via OpenRouter.

Is OpenScience free to use?+

The OpenScience software is free and open source under Apache 2.0. You pay only for the model API costs from your chosen provider (using your own keys) or optionally use Ace, a pay-as-you-go wallet starting at $20 that debits at OpenRouter prices plus 2% for credit-backed models and enhanced search.

Which AI clients or platforms does OpenScience work with?+

OpenScience is a standalone agent application, not a plugin for other AI assistants. It runs its own browser-based workspace and connects directly to model providers like OpenAI, Anthropic, Google Gemini, DeepSeek, or local models.

Can I use OpenScience with local models?+

Yes, OpenScience supports local models alongside API-based frontier models. You can configure and switch between providers from the Models settings panel without changing your workflow or losing session state.

How do I install openscience?+

Open the source repository on GitHub and follow its README. openscience is a agent — MCP Agents Market links you directly to the official repo.

Is openscience free?+

openscience is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches