Spatial transcriptomics has transformed our ability to study tissue architecture at molecular resolution, yet analyzing these data demands navigating dozens of computational methods across incompatible Python and R ecosystems---forcing researchers to devote more effort to making tools function than to pursuing biological questions. We present ChatSpatial, a platform in which the LLM selects from pre-validated tool schemas rather than generating free-form code, with domain expertise embedded in schema descriptions for context-aware parameter inference. Built on the Model Context Protocol (MCP), ChatSpatial unifies 60+ methods across 15 analytical categories into a single conversational workflow spanning Python and R ecosystems. Replication of two published studies---recovering subclonal heterogeneity in ovarian cancer and tumor microenvironment organization in oral squamous cell carcinoma---and validation across seven LLM platforms demonstrate that schema-enforced orchestration yields near-deterministic reproducibility at the workflow level for multi-step spatial analyses. Beyond replication, exploratory cross-method analyses illustrate practical triangulation across independent analytical frameworks.