MMM Labs

Best path for 'same variables, different priors': exact-clone a config, then surgically apply prior updates. Preserves exclusions. Prefer this over rebuilding selections or save_configuration_selec...

create_config_variant

Create a new configuration for a dataset. A configuration stores model settings and must exist before running models. Creates a 'default' config if config_name is not specified. After categorizatio...

create_configuration

Create a new dataset within an existing project. Datasets contain the data files used for modeling. After creation, files can be uploaded via the Data Hub. ALWAYS call this tool when the user ment...

create_dataset

Create a new project in MMM Labs. Projects organize datasets and models. ALWAYS call this tool when the user mentions creating a project, including: - 'create project X' - 'new project X' - 'start...

create_project

Render a support ticket for the user to approve. Writes nothing and emails no one. CHECK THE DOCUMENTED BEHAVIOUR FIRST, SILENTLY. The Knowledge Center and your loaded product skills are the refere...

draft_support_ticket

Meta-tool: summarize model metadata, fit statistics, and channel ROI in one call. Use before get_contributions/get_predictions for narrative explanations.

explain_model_results

Budget what-if — list all scenarios for a dataset (any configuration) and load results for scenario_name or the first match. Omit config_name unless disambiguating duplicate titles.

explore_scenario

Create engineered variables (seasonality/Fourier, trends, interactions, dummies, composites) for a configuration. Persists engineered_variables.json — confirm in Data Hub UI, then regenerate model...

generate_engineered_variables

Generate model inputs (model_inputs.json) for a configuration. This must be done after data is uploaded and categorized, and before running a model. The model inputs file contains the processed dat...

generate_model_inputs

Get the actual observed values of the dependent variable (KPI) over time. Returns the raw actuals time series that the model was trained to predict. Use together with get_predictions to compare mod...

get_actuals

Read-only: which configurations no longer describe the dataset. Reports any configuration that references a column the dataset no longer has — those cannot be fitted, and the failure otherwise surf...

get_configuration_staleness

Get per-variable contributions in dependent variable (KPI) units — every variable incl. intercept and controls. Served from the cached results view when available; the response states its source an...

get_contributions

Read-only: what would replacing a dataset file change, and may it? Classifies the change as added sheets / added columns / more observations / restated values / REMOVED column, and refuses a remova...

get_dataset_refresh_preview

Show how each variable's contribution will be MEASURED for a configuration: its reference point (the level the contribution is measured from, and whose displaced level folds into the base), plus it...

get_decomposition_settings

Get sales elasticity per modeled variable: the % change in total predicted sales per 1% change in the variable's own activity (Marginal Contribution / Total Predicted Sales), sign preserved (price...

get_elasticity

Load the generated model_inputs.json for a configuration so an agent can verify the wire (priors_configurations, configurations_keys, variable_groups) or reproduce a run locally. generate_model_inp...

get_model_inputs

Get results from a completed model run (ROI, contributions, accuracy, MAPE, R²). When the outputs file includes a predictions block, also returns fit_time_series_preview with aligned dates, actual,...

get_model_results

Get model run status for a dataset (completed, running, failed, etc.). After start_model, poll with get_model_status or wait_for_model_completion.

get_model_status

Get model parameters with posterior/prior distributions (Bayesian) or point estimates (non-Bayesian). Includes coefficients, adstock params, and saturation params for each variable. Use for underst...

get_parameters

Robyn Pareto front from the model artifact: selected solution, front candidate ids with fit metrics (nrmse, decomp_rssd, mape), and the solutions cloud (same data as the Model Results Pareto chart).

get_pareto_candidates

Get the model's per-period predictions (fitted values) over a date window. Returns dates, predicted values, and KPI name. IMPORTANT — models run with a train/holdout split are fit TWICE, and this t...

get_predictions

Get platform ROI per paid channel: revenue_contribution / media_spend (revenue $ per media $), identical postprocess basis for every engine including meridian and pymc — gaps between engines are fi...

get_roi

Get actual spend (real spend column values). Pass a channel for one channel, or omit channel to get every paid channel keyed by name (under 'by_channel') — useful for spend reconciliation. Can aggr...

get_spends

Per-variable reference levels for one fitted model: the window minimum/maximum/mean/median in the variable's own units, the level the decomposition actually measures from, how much contribution the...

get_variable_reference_stats

Get dataset-level categorization status (column → Sales/Advertising/Others). This is not configuration_selections.json (per-config variable picks). Shows how columns are categorized and their metri...

get_categorization_status

List business priors configured per channel/variable in a configuration (priorMode=business). Prefer this over loading the full configuration_selections JSON.

get_channel_business_priors

Per-column statistics (null %, min/max/mean, dtypes) for a fact table, aggregated table, wide_data, or raw upload sheet.

get_column_statistics

Load configuration_selections.json for a named configuration (per-config variable picks). Not the same as categorization_results.json. Config names from the UI (e.g. 'Config 1') must match config_n...

get_configuration_selections

Get per-period contributions for all variables (not aggregated). Returns dates and per-variable time series for decomposition charts.

get_contribution_time_series

Preview uploaded data. Without sheet_name: returns sheet metadata from data_frames.json (columns, row counts). With sheet_name: returns row samples matching the Data Hub preview — raw upload sheets...

get_data_preview

Get decomposition over time (per-period channel contributions and date axis) from completed model outputs. Use for 'decomposition chart', 'contribution by week', or time-series MMM decomposition.Re...

get_decomposition_time_series

Get agent-friendly metadata from a fitted model: algorithm, KPI, date range, frequency, revenue conversion type, transform family, cross-section info, and variable_scaling (per-channel max/max-abs...

get_model_metadata

Get model fit statistics. Returns in-sample (full/train) and out-of-sample (holdout) metrics when dataset_metrics is present, plus test_split metadata (num_obs_test, test_start). Flat r_squared/map...

get_model_statistics

Get a variable's modeling-units time series (transformed values used in the model). Useful for understanding how raw data was transformed.

get_modeling_units

Get SaaS-wide default priors configuration (adstock, saturation, coefficient constraints). Same data as GET /api/priors-defaults.

get_priors_defaults

Return the project/dataset/config pinned for this MCP session plus server-resolved workspace facts (model statuses, configurations, champion). Auto-populated from your last app selection on connect when available.

get_workspace_context

Get details of a specific project including its datasets, configurations, and metadata. Prefer this over list_datasets. Pass account_id or account_name from the list_projects row when the project i...

get_project

Get budget-scaling response curves for paid channels in REAL spend dollars (x = spend, revenue/dependent = contribution, operating point marked), plus per-channel ROI/MROI. Served from the cached r...

get_response_curves

Get the current output mode for this session. Returns the active mode and available options.

get_output_mode

Return paired x/y series ready for scatter charts. chart_type=actual_vs_predicted (default), spend_vs_kpi (requires channel), or response_curve (requires channel).

get_scatter_data