Privacy Policy
Forecast Coach ("we", "us", "our") is a time-series forecasting MCP App that AI assistants such as ChatGPT, Claude and Copilot call on your behalf. This policy covers the free public connector at https://forecast-coach.us-west.pmcp.run/mcp, its chart widget, and this website. It explains what data we receive, why, who else handles it, how long it is kept, and what control you have.
Summary
- No account and no login. The free connector is anonymous. We never ask for your name, email, password or payment details.
- Stateless. The numbers your AI assistant sends are held in memory only while the forecast is computed, then discarded. Nothing is written to a database, file or disk.
- Request contents are never logged. Our logs record only technical metadata (status code, response size, response time). They never contain your dates, values, titles or any other tool argument.
- No selling, no advertising, no model training. Your data is never sold, shared with advertisers, or used to train any AI model.
1. Data We Receive
1.1 Tool inputs
Forecast Coach exposes two tools, forecast_series and backtest. When you ask your AI assistant for a forecast, the assistant reads your spreadsheet or message itself and sends us only the fields below. We never receive the original file, your conversation, or your identity.
- The series: a list of dates (
ds) and the matching numeric values (y), for example daily sales. - Forecast settings: the horizon (number of future periods), the frequency (daily, weekly or monthly), the chosen algorithm or list of algorithms, and for a backtest the number of test origins.
- Calendar information: weekdays and dates on which the business was closed.
- Optional drivers and events: names and values of external drivers (such as a price or a promotion) and names and dates of holidays or events.
- Optional title: a short label for the chart heading.
These fields are designed for business figures, not personal information. Please do not put personal data (such as a customer's name) into a series title or a driver name; if you do, it is handled exactly like every other input: used for the one response and then discarded.
1.2 Tool outputs
We return the forecast values with their confidence band, the accuracy of each algorithm on your series, coaching notes, and the data the chart widget needs to draw them (which includes your history series). The output goes back to your AI assistant and the widget inside it. We keep no copy.
1.3 Operational metadata
To keep the service running, our server writes a small number of log lines and performance traces: one line per response containing the HTTP status code, the response size in bytes and the response time in milliseconds; start-up events; and server error messages that contain nothing derived from a request. Our hosting platform also records standard execution records (a random request ID, duration and memory used). Our logs do not contain tool arguments, tool outputs, IP addresses or request headers.
Like any internet service, the network infrastructure that delivers requests to us (see section 3) necessarily processes connection data such as the IP address of the connecting AI provider in order to route traffic and protect the service against abuse.
Country-level usage statistics. The pmcp.run routing service looks up the country a request comes from based on its IP address and increments an aggregate request counter for that country. Only these per-country counts are stored; they are not linked to any individual, conversation or tool input.
1.4 This website
This website loads one advertising-measurement script, the OpenAI Ads pixel (served from bzrcdn.openai.com), so we can measure visits that arrive from ads shown in ChatGPT. It may set a cookie or identifier in your browser; OpenAI processes that data under OpenAI's privacy policy. The website runs no other analytics or tracking scripts. It loads the Inter font from Google Fonts, so your browser contacts Google's servers, which receive your IP address under Google's privacy policy. The chart widget shown inside your AI assistant makes no network requests of its own and stores nothing in your browser.
1.5 What we do not collect
We do not collect names, email addresses, account credentials, payment information, location, contact lists, device identifiers, cookies other than any set by the OpenAI Ads pixel described in section 1.4, conversation history or the files you upload to your AI assistant.
2. How We Use Data
- Tool inputs are used for one purpose only: to compute the forecast or backtest you asked for and return it.
- Operational metadata is used to monitor availability and performance, to diagnose errors, and to detect and prevent abuse of the free service.
- Country-level counts are used to show the service administrator a map of where usage comes from, and, if ever needed, to block requests from countries on a blocklist (for example to comply with sanctions law).
We do not use any data for advertising, profiling, selling, or training AI or machine-learning models. The one exception is the advertising measurement on this website described in section 1.4, which tells us whether a visit came from an ad. The forecasting models run inside our server and are never updated from your data.
3. Who Receives Data
- Your AI assistant provider (for example OpenAI for ChatGPT, Anthropic for Claude, or Microsoft for Copilot) sends us the tool inputs and receives the tool outputs. Your conversation and uploaded files are governed by that provider's own privacy policy.
- Hosting infrastructure: the connector runs on pmcp.run, which operates on Amazon Web Services (AWS) in the United States. Requests pass through AWS API Gateway (with no access logging enabled) and the pmcp.run routing service to our server on AWS Lambda; operational metadata is stored in AWS CloudWatch and AWS X-Ray. These providers act as our service providers and process data only to run the service.
- Google Fonts, for visitors to this website only, as described in section 1.4.
- OpenAI (the OpenAI Ads pixel), for visitors to this website only, as described in section 1.4.
We do not sell, rent or share data with any other third party. We may disclose information if required by law, but because we keep no tool inputs or outputs, there is no stored business data to disclose.
4. Retention
- Tool inputs and outputs: held in memory only for the duration of the request (typically well under a second) and discarded when the response is returned. Never stored.
- Operational logs: our server's logs are deleted automatically after 30 days; the pmcp.run routing service's logs after 14 days.
- Country-level counts: kept as aggregate statistics for as long as the service operates. They contain no IP addresses or other personal information.
- Performance traces: deleted automatically after 30 days (the AWS X-Ray retention period).
5. Your Choices and Controls
- You decide what is sent. Forecast Coach is called only when you ask your AI assistant for a forecast, and the assistant shows the tool call and the data it sends. Only the numbers needed for the forecast are sent, and you can ask the assistant to leave anything out.
- Disconnect at any time. Remove or disable the Forecast Coach connector or app in your AI assistant's settings and no further data will reach us.
- Your conversation history is controlled in your AI assistant, where you can delete it.
- Access, correction and deletion. Because there is no account and we keep no tool inputs or outputs, there is no personal data held by us to access, correct, export or delete. You are still welcome to contact us with any request or question, including under the GDPR, the UK GDPR or the CCPA/CPRA, and we will respond within 30 days.
6. Security
All traffic to the public connector is encrypted in transit with TLS. Each request is processed in an isolated, short-lived execution environment with no persistent storage for request data. The connector needs no credentials, so there are none to leak.
7. Children
Forecast Coach is a business tool and is not directed to children under 13 (or the equivalent minimum age in your country). We do not knowingly collect personal information from children.
8. Private Enterprise Deployments
Organizations can run Forecast Coach inside their own AWS, Azure, Google Cloud or on-premises environment. In a private deployment the server runs in the customer's own infrastructure, data stays within its network, and data handling is governed by the customer's own policies and the separate agreement with us. This policy applies to the free public connector and this website.
9. Changes to This Policy
If our data practices change, we will update this page and the "Last updated" date before the change takes effect. We will not begin storing tool inputs or outputs without first updating this policy.
10. Contact Us
For questions or requests about privacy, contact: privacy@mlguy.us