The ART MCP server enables full Box-Jenkins-Treadway univariate time series analysis, driven by an LLM in either guided or autonomous mode. Key capabilities include:
Data Ingestion & Preparation: preview_data (preview Excel/CSV), load_data (load and create .inp), create_inp (from raw data), series_info (get metadata).
Exploratory Analysis: boxcox_analysis, unit_root_analysis (ADF, KPSS), seasonal_analysis (HAC F-test), identification_analysis (ACF/PACF with ARMA suggestions), preliminary_outlier_scan (detect outliers in differenced series).
Model Identification: guided_identification (step-by-step sequential decisions on Box-Cox, differencing, seasonality, ARMA orders).
Model Estimation & Diagnosis: estimate_and_diagnose, confirm_and_estimate, model_histogram, overparameterization_analysis (parameter correlations), model_equation_display (polynomial operators), formal_tests (Shin-Fuller, DCD, RV, MEG), ar_factorization (factor AR operators).
Seasonality & Interventions: seasonal_param_analysis (visualize seasonal harmonics), test_seasonal_simplification (joint LR test), intervention_analysis (detect extreme residuals and assess impact), test_interventions (test significance), suggest_intervention_form (add pulse/step/ramp), meg_reformulate, meg_frequency (stochastic seasonality).
Automated Modeling: build_model (full pipeline, autonomous or guided), batch_build (multiple series).
Versioning & Audit Trail: record_version (log model in guion.json), compare_versions (spec diff, stats, LR test), export_guion (HTML audit report).
Reporting: full_report (self-contained HTML diagnosis report), save_identification_report (HTML identification report), get_out_report (full ASCII output).
Forecasting: generate_forecast (L-step forecasts, HTML report), update_and_forecast (update with new data, tracking errors), sps_dashboard (sequential prediction dashboard for multiple series).
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@ART MCP Server Run autonomous ARIMA identification on my monthly sales data"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
art-tseries (ART) builds univariate time series models following the
Box-Jenkins-Treadway methodology: an iterative, decision-driven process that
uses graphical tools and formal tests to identify, estimate, diagnose and refine
a model until it is adequate and parsimonious.
ART is the orchestration layer of a four-part suite:
Package | Role |
Exact maximum-likelihood estimation (ARMAX + transfer functions) and FUF forecasting. C engine with a pure-Python fallback. | |
pyfug | High-definition graphics for time series analysis. |
ART ( | Identification, model building, diagnosis, formal tests, versioning — and an MCP server that exposes all of this to an LLM. |
The Box-Jenkins-Treadway loop needs judgement at each decision node. ART supplies the evidence (graphs, tests, numbers); a human analyst and/or Claude supply the criterion. Two modes:
Guided — analyst + Claude: Claude proposes with arguments, the analyst decides.
Autonomous — Claude/heuristic decides every step and presents a final model.
pip install art-tseries # pulls fue + pyfug automaticallyThis installs the art-mcp command (the MCP server).
Related MCP server: Prophet MCP Server
claude mcp add art -- art-mcpThen ask Claude to analyse a series. ART will ask whether you want a guided or autonomous analysis and drive the workflow from there.
import fue
from art.describe import describe_boxcox, describe_identification, model_equation
ts, _ = fue.inp.load("series.inp")
print(describe_boxcox(ts).summary)The model-building process is iterative and sequential: each estimation starts
from the previous likelihood optimum (the .pre of the previous model), and
every step produces a .pre (estimated parameters as initial values) and a
.out (results), mirroring fue. Decisions and changes are recorded in a
guion.json audit trail. See docs/ARCHITECTURE.md for the full design and the
evidence-vs-criterion philosophy.
PyPI renders this page only, so everything below is linked absolutely and also
ships inside the source distribution — pip download art-tseries --no-binary :all: — so it reads without network.
install, first analysis, EN / ES | |
every | |
how ART, FUE, FUG and FUF fit together | |
why the scale factor exists and where it acts | |
what changed, version by version |
The suite as a whole installs with pip install atsw.
GPL-2.0-or-later. © David E. Guerrero.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
The statistical analyst in your AI chat — validated, citable, re-runnable analysis of your data.
Agent-to-agent reasoning-as-a-service: chain-of-thought, analysis, and decision support.
Wall-clock awareness for LLM agents. Two tools: elapsed-time-between-turns + day rollover detection.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/davidesg/art-python'
If you have feedback or need assistance with the MCP directory API, please join our Discord server