AutoTS MCP
AutoTS MCP server with 28 tools: an agent loads series, forecasts, finds anomalies and rates threshold-crossing odds.
- 1,433 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Reads CSV files by path and by URL and calls external services when loading live data.
Install
- Works in
Manual install
pip install autots[mcp]The first command from the README, installs the library with MCP dependencies.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Reads CSV files by path and by URL and calls external services when loading live data.
This is third-party code. Review the repository files before installing.
What it does
AutoTS is an automated time series forecasting library for Python. Its MCP server starts with the `autots-mcp` command over stdio and follows a cache-and-ID pattern: each operation stores its result on the server and returns an ID that is passed to the next tool. Tools are grouped: data loading (built-in sets, CSV by path or URL, live FRED and stock data, synthetic series), cleaning, three forecast modes and result handling. `forecast_fast` builds a ready ensemble in seconds, `forecast_explainable` searches only interpretable models and `forecast_custom` accepts any AutoTS parameters. Then come plots, constraints and adjustments to a forecast, decomposition into trend and seasonality, the probability of crossing a threshold (stockout, capacity breach) and detection of anomalies, level shifts and holidays.
Who it is for. Analysts and data engineers who forecast demand, load or metrics and want to run the calculation through an agent with no hand-written code.
Good fit when
- An agent loads a sales CSV and builds a quick forecast a month ahead in seconds.
- An analyst asks for an explainable model and looks at the trend and seasonality decomposition.
- A planner estimates the probability that warehouse stock falls below a threshold.
- A metric series needs anomalies, level shifts and holiday effects found in it.
Not a fit when
- You need a deep neural model for one series, which the ready modes do not replace.
- The data is confidential while the server calls external sources such as FRED.
- The series do not fit in the memory of the machine running the server.
Example request
Load the file sales.csv, build a quick 30-day forecast and show a plot with intervals.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Install with `pip install autots[mcp]` or `pip install autots-mcp`. Results live in the process cache, so IDs are lost when the server restarts and `list_cache` shows the live ones. Live FRED and stock data need network access. `forecast_custom` with a full model search takes a long time and `get_autots_docs` should be called before it. The `forecast_from_features` method is marked experimental. Data is expected in wide format, and long format is converted automatically.
How to disable. Remove the autots entry from the MCP configuration and run `pip uninstall autots-mcp`. The cache lives in process memory and disappears when the server stops.
FAQ
How do I connect AutoTS to VS Code?
Add an `autots` server with the command `autots-mcp` to `.vscode/mcp.json`. For a local checkout run `python -m autots.mcp.server` with the AutoTS working directory.
How do the three forecast modes differ?
`forecast_fast` takes a ready ensemble and runs in seconds, `forecast_explainable` searches among interpretable models and `forecast_custom` runs the full AutoTS search with your parameters or a template.
What are the IDs in tool responses?
Each operation stores its result in the server cache and returns an identifier: `data_id` for data and `prediction_id` for a forecast. It is passed to the next tool, and `list_cache` lists all active ones.
Can I load data from FRED?
Yes, `load_live_data` pulls data from the FRED economic API or stock tickers. It needs network access.
How do I estimate the probability of crossing a threshold?
Call `forecast_event_risk` with the threshold, then `get_event_risk_results` and `plot_event_risk`. This is how stockout or capacity breach is estimated.
Discussion
No questions or reviews yet.
Related
MCP serversA Python bootstrap library for time series with an MCP server: series diagnostics and a confidence interval without writing code
MCP server for Open-Meteo: 17 tools, a 16-day forecast, archive from 1940, air quality, marine data, floods and climate
An MCP server with keyless weather: forecasts, alerts, air quality, lightning, rivers, wildfires and history from 1940.
Official Apache IoTDB MCP server: SQL over time series in tree and table models, with CSV and Excel export
Roboto MCP: robotics data for AI agents
Roboto MCP
The hosted Roboto MCP server lets an agent search a robot fleet's datasets and files and analyze topic data without changing them
An MCP server with a toolbox for global briefs: markets, SEC filings, conflict, climate and news from open sources
README in short
The server lives in the `autots.mcp` package and ships with AutoTS when installed with the `mcp` extra. The package README describes twenty-eight tools and typical call chains.