AutoTS MCP Server: Time Series Forecasting for Claude, Cursor and Other Apps

• Lists saved data and forecasts

  Shows every dataset, forecast, model and detector the server is holding in memory, with the IDs other tools need.

  Try: “What data and forecasts are saved right now?”

  Read · list_cache

• Clears saved results

  Removes chosen objects, one type of cache or everything from the server's memory.

  Try: “Clear all saved forecasts.”

  Write · clear_cache

• Loads a built-in dataset

  Loads one of the example datasets, such as daily, hourly, weekly, monthly, yearly, linear, sine or artificial.

  Try: “Load the monthly sample data.”

  Read · load_sample_data

• Loads FRED or stock data

  Pulls live economic series from FRED or stock prices by ticker, over the internet.

  Try: “Load the last 5 years of AAPL and MSFT prices.”

  Network · load_live_data

• Makes test series

  Generates daily series with labeled trend, seasonality and noise parts, for testing.

  Try: “Make a synthetic daily series with weekly seasonality.”

  Read · generate_synthetic_data

• Loads a CSV file or link

  Reads a CSV from a local path or a web address. The first column must be the date.

  Try: “Load /Users/me/sales.csv.”

  Files · load_data_from_file

• Loads pasted or uploaded data

  Accepts pasted CSV or TSV text, a CSV link such as a published Google Sheet, or an uploaded CSV or XLSX. It removes empty rows and columns and works out wide versus long layout.

  Try: “Here is my sales table, load it.”

  Network · smart_load

• Returns saved data

  Returns saved data as JSON in wide or long form, or writes it to a CSV file.

  Try: “Save the loaded data as sales_clean.csv.”

  Write · get_data

• Reshapes data

  Converts data with date, series ID and value columns into one column per series.

  Try: “Convert my long data to wide format.”

  Read · convert_long_to_wide

• Fills gaps and outliers

  Fills missing values and handles outliers using methods such as forward fill, mean, median, rolling mean or linear.

  Try: “Fill missing values with linear interpolation.”

  Read · clean_data

• Makes a quick forecast

  Runs a pre-set ensemble with no model search, so results come back in seconds. It returns an ID for the forecast.

  Try: “Forecast the next 30 days quickly.”

  Read · forecast_fast

• Forecasts with readable models

  Searches only models that explain themselves (Cassandra, TVVAR, BasicLinearModel), so you can see trend and seasonality.

  Try: “Give me an explainable forecast I can break into parts.”

  Read · forecast_explainable

• Forecasts with your own settings

  Runs the full AutoTS search with the parameters or model template you give. It takes longer than the quick option.

  Try: “Run a full model search with a 5-minute budget.”

  Read · forecast_custom

• Returns forecast numbers

  Returns the point forecast and upper and lower bounds as JSON or CSV.

  Try: “Show me the forecast and its intervals.”

  Read · get_forecast

• Draws the forecast

  Plots the forecast with history and prediction intervals and returns the image.

  Try: “Plot the forecast for series1.”

  Read · plot_forecast

• Limits a forecast

  Dampens growth, sets upper or lower bounds or clips by quantile, and saves the result as a new forecast.

  Try: “Cap the forecast at 1000 and floor it at 0.”

  Read · apply_constraints

• Nudges a forecast

  Applies a linear ramp, aligns to the last value or smooths with an EWMA, and saves a new forecast.

  Try: “Align the forecast start to the last actual value.”

  Read · apply_adjustments

• Shows which model made a forecast

  Returns the model name, its settings and the transformations used for a forecast.

  Try: “Which model produced this forecast?”

  Read · get_model_params

• Splits a forecast into parts

  Returns trend, seasonality and other components, for Cassandra and TVVAR models only.

  Try: “Break the forecast into trend and seasonality.”

  Read · get_forecast_components

• Shows model rankings

  Returns cross-validation results and the top models from a model search.

  Try: “Which models won the search?”

  Read · get_validation_results

• Draws validation forecasts

  Plots the cross-validation forecasts from a model search and returns the image.

  Try: “Plot the validation results.”

  Read · plot_validation

• Estimates the chance of crossing a limit

  Estimates the probability that a series passes a threshold in coming periods, such as a stockout or capacity breach.

  Try: “What is the chance demand exceeds 500 next month?”

  Read · forecast_event_risk

• Returns event risk numbers

  Returns event risk probabilities as JSON or CSV.

  Try: “Show the event risk probabilities.”

  Read · get_event_risk_results

• Draws event risk

  Plots the event probabilities across the forecast horizon.

  Try: “Plot the event risk.”

  Read · plot_event_risk

• Finds anomalies and patterns

  Detects anomalies, changepoints, level shifts, holidays and seasonality in a series.

  Try: “Find anomalies and level shifts in this series.”

  Read · detect_features

• Queries what was found

  Returns detected features, optionally limited to a date range or series.

  Try: “Show anomalies from last quarter only.”

  Read · get_detected_features

• Draws the detected features

  Plots detected features on top of the time series.

  Try: “Plot the detected features.”

  Read · plot_features

• Forecasts from detected parts

  Experimental. Builds a forecast from the decomposed parts that feature detection found.

  Try: “Forecast from the detected components.”

  Read · forecast_from_features

Watch out for

• It reads local files and fetches web addresses

  load_data_from_file takes a local path or URL, and smart_load takes a URL. Only point it at data you mean to share with the assistant. github.com

• get_data can write a CSV file

  Its save option writes a CSV to the path you give, which can replace an existing file. github.com

• Big searches can crash on low memory

  The README says memory shortage is the most common cause of random crashes. Try a data subset or a smaller model list. github.com

Security advisories