Glossary#

Technical terms and concepts used throughout ToolUniverse documentation.

AI Scientist#

An AI system (LLM, agent, or reasoning model) enhanced with ToolUniverse capabilities to autonomously perform scientific research tasks including literature search, data analysis, and experimental design.

AI-Tool Interaction Protocol#

Standardized interface governing how AI scientists issue tool requests and receive structured results. Ensures compatibility across different LLM providers and tool implementations.

Agentic Tool#

A tool that uses LLM capabilities internally to perform complex tasks like summarization, extraction, or intelligent search. Examples: Tool_Finder_LLM, summarization hooks.

ChEMBL#

European Molecular Biology Laboratory’s database of bioactive drug-like small molecules. ChEMBL IDs are identifiers like “CHEMBL25” for aspirin.

ChEMBL ID#

Unique identifier for chemical compounds in the ChEMBL database (e.g., CHEMBL25 for aspirin).

Compact Mode#

Context window optimization mode that exposes only 4-5 core discovery tools instead of 1000+ tools. Reduces context usage by 99% while maintaining full functionality through the execute_tool tool.

EFO#
EFO ID#

Experimental Factor Ontology - A standardized disease classification system used in biomedical research. EFO IDs are identifiers like “EFO_0000537” for hypertension.

Embedding#

Vector-based semantic search where text is converted to numerical vectors. Similar concepts have similar vectors, enabling “meaning-based” search beyond keyword matching. Used in Tool_Finder for intelligent tool discovery.

Hook#

Output processing function that transforms or enhances tool results. Examples: SummarizationHook (AI-powered summaries), FileSaveHook (automatic file saving).

MCP#
Model Context Protocol#

Open standard protocol enabling AI assistants to securely connect to external tools and data sources. Think of it as “USB for AI tools” - a standardized connection interface.

SMCP#
Scientific Model Context Protocol#

ToolUniverse’s implementation of MCP, optimized for scientific workflows. Extends standard MCP with scientific domain features, intelligent tool discovery, and hooks system.

Profile#
Toolspace#

YAML configuration file (profile.yaml) declaring which tools to load, cache settings, LLM config, and hooks. Profiles can be loaded from GitHub, local files, or URLs using --load parameter.

STDIO Transport#

Communication method where data flows through standard input/output streams. Used by desktop applications like Claude Desktop to communicate with MCP servers.

Tool Composition#

Chaining multiple tools together in sequential or parallel workflows. Example: Search PubMed → Extract abstracts → Summarize findings.

Tool Finder#

Family of specialized search tools for discovering relevant ToolUniverse tools:

  • Tool_Finder_Keyword: Fast keyword search

  • Tool_Finder_LLM: Natural language search using LLM

  • Tool_Finder: Semantic embedding search (most powerful)

Tool Specification#

JSON definition describing a tool’s purpose, input parameters, output format, and execution details. Think of it as the “instruction manual” for each tool.

ToolSpace#

See Profile.

PubChem CID#

PubChem Compound Identifier - unique numerical ID for chemical compounds in PubChem database (e.g., CID 2244 for aspirin).

UniProt Accession#

Unique identifier for protein sequences in UniProt database. Format: 1-6 characters (e.g., “P05067” for amyloid precursor protein).

FAERS#

FDA Adverse Event Reporting System - Database of adverse drug reactions reported to FDA. Used for pharmacovigilance and drug safety monitoring.

SSE#

Server-Sent Events - HTTP-based protocol for servers to push updates to clients. One of the transport methods supported by ToolUniverse MCP servers.

TTL#

Time To Live - Duration (in seconds) that cached results remain valid before expiration. Set via TOOLUNIVERSE_CACHE_DEFAULT_TTL.

LRU Cache#

Least Recently Used cache - Memory cache that evicts oldest unused entries when full. ToolUniverse uses LRU for in-memory caching.

Rate Limit#

Maximum number of API requests allowed per time period. Example: NCBI allows 3 requests/second without API key, 10 requests/second with key.

Tool Category#
Tool Type#

Classification of tools by data source or functionality. Examples: uniprot, chembl, pubmed, fda, nvidia_nim. Used for selective tool loading.

Fingerprint#
Cache Fingerprint#

Unique hash computed from tool’s source code and parameter schema. When tool implementation changes, fingerprint changes, invalidating old cache entries.

Workspace Configuration#

YAML or JSON file defining tool selection, hooks, and settings for a specific research workflow or domain. See Profile.

JSON-RPC#

Remote Procedure Call protocol using JSON for data encoding. Some MCP servers use JSON-RPC for communication.

OpenTargets#

Open Targets Platform - Database connecting genes to diseases with evidence scores. Provides target-disease associations for drug discovery.

AlphaFold#
AlphaFold2#

AI system for predicting 3D protein structures from amino acid sequences. Accessible through ToolUniverse via NVIDIA NIM tools.

NCBI#

National Center for Biotechnology Information - US government agency providing biomedical databases including PubMed, GenBank, and protein databases.

Ontology#

Structured vocabulary defining terms and relationships in a domain. Examples: EFO (diseases), Gene Ontology (biological processes), ChEBI (chemicals).

API Endpoint#

URL path for making API requests to web services. Example: /api/search or https://rest.uniprot.org/uniprotkb/search.

See Also#