Interoperability in AI Agent Systems: Protocol and Artifacts for a Multivocal Systematic Mapping Study
Erasmo Leite Monteiro · Zenodo (CERN European Organization for Nuclear Research) · 2026
This record contains the pre-registered protocol and research artifacts ofan ongoing multivocal systematic mapping study on how interoperability isconceptualized in the literature on AI agent systems and agentinteroperability protocols (MCP, A2A, ACP, ANP). Motivation. Verification on the field's canonical survey found zerooccurrences of "pragmatic", "LCIM", "semiotic" and "levels ofinteroperability", suggesting that the agent protocol literature hasdeveloped without contact with four decades of research on interoperabilitytypes, levels and models. This study maps what that literature actuallyaddresses, using the consolidated taxonomy of 36 interoperability types(Maciel, Valle, Santos & Nakagawa, ACM Computing Surveys, 2024) as areference instrument. Contents.- Study protocol (research questions, search strategy, selection criteria, classification scheme, threats to validity)- Database-specific search strings and execution log- Quasi-Gold Standard and search string validation (Zhang, Babar & Tell, 2011)- Rule-based screening assistant (source code) and its validation- Consolidated corpus of 1,028 unique records (2024-2026) retrieved from OpenAlex, Semantic Scholar, arXiv and Crossref- Prioritized screening spreadsheet- Premise test on the 25 highest-priority records Methodological note. The study is conducted by a single author withoutaccess to subscription databases (Scopus, Web of Science, IEEE Xplore, ACMDigital Library). Both constraints are declared, and the protocol adoptsthe mitigations prescribed by Kitchenham & Charters (2007) forsingle-researcher reviews: explicit and pilot-tested criteria, test-retestintra-rater reliability, and full public disclosure of every screeningdecision. This protocol is deposited BEFORE full screening, so that researchquestions and predictions are on record prior to results.