ChatSpatial is an MCP server for spatial transcriptomics analysis, providing a natural language interface to invoke analysis tools. Its core design replaces LLM ad hoc code generation with schema-enforced orchestration: instead of generating arbitrary scripts, the LLM selects tools and parameters from a curated tool registry, making spatial transcriptomics workflows more reproducible across sessions and clients.
Key capabilities ChatSpatial exposes 20 schema-validated MCP tools, orchestrating 65 spatial transcriptomics methods across 13 analysis categories. Tools are stable natural language interfaces, and methods are analysis backends selected via tool parameters. The currently covered methods support data types such as 10x Visium, Xenium, Slide-seq v2, MERFISH, and seqFISH.
Protocol and deployment The server implements MCP 2026-07-28 through the official Python SDK v2, and continues to serve 2025-11-25 clients through SDK-managed protocol negotiation. STDIO remains the secure local default; Streamable HTTP can be used for explicitly configured HTTP deployments.
Installation and usage Install uv once, then register ChatSpatial without creating or managing a Python environment. uvx creates an isolated environment on first launch and reuses its cache on subsequent launches; after adding the server, restart the MCP client. The above command installs the standard runtime. To provide all 15 composable Python method families in the same isolated MCP environment, use full mode. full includes CellRank, FastCCC, maintained spatial domain and registration backends, annotation, enrichment, and other portable Python method families. The R bridge, AESTETIK, and rctd-py remain separate because they have system, platform, or large runtime requirements; see the installation guide before enabling these method families.
Quick start and analysis workflow The documentation provides a quick start to run the first analysis: the quick start includes downloadable samples and a Leiden workflow that can run under the standard installation; its samples have spatial coordinates but no H&E images. SpaGCN requires the optional spatial-domains extra. If using Docker, you can mount host datato /data and use container paths (e.g., /data/spatial_data.h5ad) in prompts. A minimal example prompt is provided. There is also an installation guide for selecting optional method families or persistent environments, a configuration guide for configuring other MCP clients, and a reproducibility image that can reproduce key manuscript results without using an LLM.
Compatible clients ChatSpatial works with any MCP-compatible client, including Claude Code, Claude Desktop, Codex, OpenCode, and other tools that support MCP.
Reproducibility Each release publishes a reproducibility image under the -repro tag. This image re-executes the DLPFC spatial domain and OSCC CARD deconvolution tool calls recorded in the manuscript experiments, without invoking any LLM, and compares the results with the reported values. The reproducibility/ directory contains the image definition and describes what it checks; the manuscript's experiments and analysis scripts are located in ChatSpatial-Reproducibility. If you use ChatSpatial in your research, cite it as required; because it orchestrates many excellent third-party methods, you should also cite the original tools used in the analysis.
Contributing and license Documentation improvements, bug reports, and new analysis methods are welcome; see CONTRIBUTING.md. The project is under the MIT license.
Limitations The R bridge, AESTETIK, and rctd-py are provided separately due to system, platform, or large runtime requirements, and the installation guide must be consulted before enabling them. SpaGCN depends on the optional spatial-domains extra. The quick start samples do not include H&E images. This document does not provide a complete list of details such as specific method names, parameters, and installation commands, and none is provided for now.