Documentation
Everything you need to use, build, and publish toolkits on SciToolkit.
What is SciToolkit?
SciToolkit is a registry of scientific tools for AI agents. A researcher writes a tool once — a stellar forward model, a particle physics event selector, a neutrino flux calculator — publishes it, and any AI agent (Claude Code, Codex, custom Orchestral agents) can discover and use it.
Toolkits run in isolated environments (virtualenv, conda, or Docker depending on their requirements), so you can have one toolkit on Python 3.12, another on Python 2.7, and a third with CUDA — all serving tools to the same agent.
Documentation pages
Quickstart →
Install the CLI, find a toolkit, and start serving it to your AI agent in under five minutes.
Scaffold with an agent →
Hand a single prompt to your coding agent and let it scaffold, fill in, validate, and publish your toolkit. The fastest way to ship a first toolkit.
Authoring Guide →
Create, package, and publish your own toolkit. Covers metadata, tool definitions, validation, and the publish workflow.
Authentication →
How the CLI signs in. One per-user token works across every toolkit you own or collaborate on. Browser-flow and non-interactive token modes.
Configuration & Setup →
How toolkits collect configuration from users — API keys, file paths, data downloads, and per-toolkit setup scripts.
Stateful Tools →
How tools accept configuration values that the AI agent never sees — base directories, opacity paths, worker counts.
Recipes →
Copy-paste solutions for common patterns: API keys, data downloads, multi-option setup, custom validation.
Need help?
Open an issue on GitHub or email scitoolkit.dev@gmail.com.