SPSS Studio MCP for Mac

Make SPSS an Agent's "statistics engine + chart factory": paper-ready charts, deep result parsing, verified methods, and safe execution.

This MCP is Mac Only, for Windows versions, see flupke91/spss-studio-mcp.

English ｜ 繁體中文（香港）

spss-studio-mcp is an MCP (Model Context Protocol) server for IBM SPSS Statistics. It gives Agent clients such as Codex / Claude Code / Cursor:

• Paper-ready charts: 11 spss_chart_* tools export PNG / TIFF (1950×1500 @300 dpi) in one call and return the file path for direct submission to journals;

• Deep result parsing: OMS text → Markdown tables + structured JSON + statistical summaries for 16 analysis families (t / F / r / B / Wald / α / χ² / p / effect sizes);

• Verified methods: all 37 analysis tools plus 11 supporting tools are verified on real SPSS Statistics 32 for Mac;

• Mediation / moderation: spss_mediation (Baron & Kenny three-step regression + Sobel test) and spss_moderation (mean-centred interaction regression);

• Safety layer: dangerous-command blocking, data path allowlist, dry_run preflight, and JSONL audit logging.

Everything is verified on IBM SPSS Statistics 32.0.0.0 for macOS — 105 unit tests pass, including a self-contained real-machine reproduction manifest. Per-case results: docs/macos_verification.md.

Quick Start

macOS (SPSS Statistics 32 for Mac is auto-discovered inside the .app bundle under /Applications; Python ≥3.10 required):

cd spss-studio-mcp
bash scripts/install_macos.sh              # creates .venv, installs deps, configure-claude
.venv/bin/spss-studio-mcp status           # SPSS batch: OK
.venv/bin/spss-studio-mcp configure-codex  # optional: Codex config

Then drive it from your client in natural language, e.g.:

Run descriptive statistics and reliability analysis on examples/data/survey_study.sav
Independent-samples t-test on examples/data/experiment_study.sav (grouping: group, dependent: posttest)
Mediation analysis on examples/data/mediation_study.sav: autonomy → satisfaction → performance
Histogram with normal density of engagement_total in examples/data/survey_study.sav, PNG 300dpi

Related MCP server: mcp-macos-cua

Paper-Ready Charts (Core Feature)

spss_chart_histogram_density(
    variable="engagement_total",
    title="Engagement total distribution (with normal density)",
    image_format="PNG",            # PNG / TIFF
    width_px=1950, height_px=1500, dpi=300,
    data_file="examples/data/survey_study.sav",
)
# → returns the image file path, ready for submission

Format note: chart output is PNG and TIFF only. Windows vector EMF export was removed because SPSS for macOS cannot produce EMF metafiles (its OMS FORMAT=DOC archive contains only raster PNG wrapped in .eps). Use image_format="PNG" or "TIFF".

Structured Results & Statistical Summaries

spss_structured_result(
    syntax="T-TEST GROUPS=group(1 2) /VARIABLES=posttest.",
    data_file="examples/data/experiment_study.sav",
)
# → {markdown, json: {tables, summary}, files, warnings}

The Markdown returned by run_syntax ends with a ### Statistical Summary block containing a plain-language conclusion plus key statistics. Extraction details for the 16 analysis families: docs/result_parsing.md.

Sample Datasets

examples/data/ ships five research-style datasets (fixed random seeds, fully reproducible):

Safe Execution

• Dangerous commands (HOST / ERASE / DELETE FILE, ...) are blocked at the start of a line;

• Data files must live under examples/, the system temp dir, or a directory declared in SPSS_ALLOWED_DIRS;

• spss_run_syntax(..., dry_run=True) validates without executing;

• Audit logs are written to logs/audit.jsonl by default (override with SPSS_AUDIT_LOG).

See docs/security.md.

Documentation

• Tutorial

• Technical report

• Chart pipeline PoC & SPSS 32 findings

• Result parsing (16 analysis families)

• Method verification

• macOS verification report (SPSS 32 for Mac)

• Security layer

• Changelog

License

MIT (upstream: flupke91/spss-studio-mcp, MIT).

MIT (upstream: Exekiel179/SPSS-MCP, MIT).