The second part of the project: https://doi.org/10.5281/zenodo.18247208 The third part of the project: https://doi.org/10.5281/zenodo.18247371 This paper presents the first part of a two-article study proposing a formal, token-based framework for analysing the manuscript independently of natural-language assumptions. Rather than treating Voynichese as an encrypted or unknown language, the present work approaches the text as a structured system of operational tokens governing process execution. A complete inventory of recurrent EVA tokens is constructed using the Zandbergen transliteration standard. Tokens are analysed as compositional units consisting of a stable core and optional modifiers. These modifiers systematically alter execution mode, duration, intensity, or closure of an operation without changing its functional class. Special attention is given to positional and allomorphic variants, including closure markers, regime modifiers, and composite operators. The study demonstrates that token variants exhibit consistent positional behaviour across distinct sections of the manuscript, including botanical, balneological, astronomical, and recipe-like folios. This invariance supports an algorithmic rather than lexical interpretation of the text. The resulting operational lexicon provides a reversible and testable formal model that can be applied uniformly across the manuscript without recourse to external linguistic decoding. This first part establishes the formal grammar and operational vocabulary of the Voynich Manuscript. A subsequent paper applies this lexicon to a functional interpretation of the manuscript as a procedural guide governing transformation processes under controlled regimes, temporal constraints, and stabilization phases.