ToolUniverse is a collection of tools, interfaces, and supporting components for building AI systems that perform scientific work. It is for developers creating AI scientist agents that use APIs, databases, machine-learning tools, and domain-specific utilities. The catalogue includes skills, commands, an MCP server, an agent, and a hook for working with the ecosystem.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/mims-harvard/tooluniverse/researchgit clone --depth 1 https://github.com/mims-harvard/ToolUniverseWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/mims-harvard/tooluniverse/research)<a href="https://agentmods.dev/commands/mims-harvard/tooluniverse/research"><img src="https://agentmods.dev/badge/commands/mims-harvard/tooluniverse/research.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00099 | $0.01198 |
| Opus 5 | $0.00049 | $0.00599 |
| Sonnet 5 | $0.00020 | $0.00240 |
| Haiku 4.5 | $0.00010 | $0.00120 |
Grade A, and why
research scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
2. **Execute**: call tools or skills directly. For batches of 5+ similar queries, use the Python SDK pattern (see below) — `subprocess.run(["tu", "run", ...])` per call reloads the registry every time. How it starts
The opening of the file, as written. The whole thing — 76 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Research the following using ToolUniverse: $ARGUMENTS
Operating Rules
-
Look up specific claims; don't guess them. For exact counts, p-values, database records, drug approvals, variant annotations, pathway memberships — call a tool. Memory is only for textbook-level background.
-
Don't search for tools you already have. For the common tools in the "Anchor Tools" section below, call them directly. For anything else, run
find_tools("topic")once, thenget_tool_info(name)once. If those two steps don't surface a usable tool, stop searching and answer from sources you do have. -
Cross-validate high-stakes facts. For claims someone will publish, cite, or act on, check at least 2 independent databases (different upstream maintainers, not different tools wrapping the same API). State agreement or disagreement explicitly.
-
An honest "INDETERMINATE" beats a confident wrong answer. If the question can't be settled from available sources, say so and explain what's missing. Don't fill gaps with plausible-sounding inferences the user may mistake for verified facts.
-
For data-file analysis (CSV, TSV, h5ad, VCF, FASTA, tree files, etc.), route to the matching ToolUniverse skill before writing code:
Skill('tooluniverse') # loads the router, which dispatches to the specialized skillThe skills encode discovered analysis conventions (which library, which filter, which denominator) that ad-hoc code often gets wrong. Do not re-derive conventions inside this chat.
Workflow
- Plan: state which tools or skills you'll need before calling them.
- Execute: call tools or skills directly. For batches of 5+ similar queries, use the Python SDK pattern (see below) —
subprocess.run(["tu", "run", ...])per call reloads the registry every time. - Analyze: when post-processing is needed, write Python via Bash and read the output.
- Validate: cross-check key findings against a second source.
- Report: actual numbers, source databases cited, dates retrieved, and an INDETERMINATE verdict where the evidence won't settle.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 76 lines · 99 tokens per session scan A bce519e16140
research is a command published in the GitHub repository mims-harvard/ToolUniverse (1,672 stars, last pushed today), licensed Apache-2.0. It adds 99 tokens to every session and 1,198 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 1 finding (runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
writing
You are the Writing Agent for NanoResearch. Your job is to generate publication-quality figures and write a complete LaTeX research paper.
experiment
You are the Experiment Agent for NanoResearch. This command combines the Setup, Coding, and Execution stages. You will generate runnable experiment code and execute it.
astro-dso-doc
Generate pixinsight project information for a deep-sky object astrophotography session.
ace-strategies
Inspect ACE skillbook strategies for a MedSci domain session.
checklist
Generate a custom checklist for the current feature based on user requirements.
clarify
Identify underspecified areas in the current feature spec by asking up to 5 highly targeted clarification questions and encoding answers back into the spec.