Agentic SWMM: Auditable and Reproducible Stormwater Modelling Workflow with Skills and Model Context Protocol

Zhonghao Zhang , Caterina Valeo

Configuring urban hydrological models, such as the Storm Water Management Model (SWMM), for operational use remains onerous for many modellers. We propose aiswmm, a SWMM-specialized Agentic runtime, together with an Agentic SWMM workflow that embeds Skills and Model Context Protocol (MCP) tools to automate GIS preprocessing, model configuration, execution, and postprocessing. Each Skill encodes a hydrologic standard operating procedure (parameter defaults, validation thresholds, metric-extraction rules), narrowing the large language model's (LLM) role from open-ended generation to constrained Skill selection. To ensure that the entire Agentic SWMM workflow is auditable and reproducible, we proposed a verification-first, provenance contract embedded into Agentic SWMM workflow, that enables byte-level audit chain to LLM-driven hydrological modelling. Each run will create an independent manifest file that documents the SWMM version used, parameter mappings, input file hashes, and quality gate (e.g., continuity diagnostics). We demonstrate the workflow on the Tod Creek watershed (located on the Saanich Peninsula, British Columbia). We validated the proposed Agentic SWMM workflow at three levels: i) a QGIS based watershed-pour-point detection that agrees with the commercial PCSWMM® method to within 0.88% of the watershed perimeter (~7.5 pixels in the digital elevation model); ii) byte-identical SWMM output files (Secure Hash Algorithm 256-bit identical) between the command-line execution and the MCP paths across 60 paired simulations, and iii) peak inflow at the watershed outlet matching to three significant digits between the manual SWMM interface and Agentic SWMM workflows. The aiswmm runtime, Skills, MCP servers, and byte-level audit chain are released as open source and remain compatible with mainstream agentic runtimes (Codex, Claude Code, Hermes, and OpenClaw) to support reproducible SWMM modelling driven by natural language.