Skip to Content
Fluxmail

Telemetry

Turn off anonymous telemetry and see what usage data and optional feedback Fluxmail sends.

Fluxmail collects anonymous usage events and error reports by default. You can turn them off without disabling email operations or feedback you explicitly choose to send.

Turn telemetry off

fluxmail telemetry disable

You can also set FLUXMAIL_TELEMETRY=0 or DO_NOT_TRACK=1. Any disabling source takes priority over an enabling source. Use fluxmail telemetry status to check the effective state. These settings disable both usage events and error reports.

Collected data

Events include a random installation ID, the Fluxmail and Node.js versions, operating system and architecture. They identify the installation as self-hosted. Operation events record the CLI command, MCP tool or REST operation, its outcome, duration and selected feature modes. CLI events also note when a command used a Fluxmail Cloud profile.

Depending on the operation, events can also include:

  • Provider diagnostics: Gmail, Outlook or IMAP, the step that failed (such as connecting, searching or refreshing a token), a fixed failure category (such as a refused connection, timeout or missing Gmail permission), the provider's HTTP status and a standard network error code such as ECONNREFUSED. Internal, provider, timeout and uncertain-send failures also produce error reports with the operation name and a safe error code. Partial account or item failures count as errors; batch search diagnostics describe the first failed account group.
  • Step timings: for MCP and REST operations on an IMAP mailbox, how many milliseconds went to connecting, listing folders, opening a folder, searching, fetching messages and other commands. Steps that run at the same time add up, so a batch search can report more time than it took.
  • Search diagnostics: when Fluxmail rejects a search query, the fixed reason, such as an invalid date or an unsupported folder. The query text is not sent.
  • Rejected tool arguments: when an MCP tool call doesn't match the tool's input schema, the name of the first argument Fluxmail rejected, a fixed reason such as missing or wrong_type, and how many problems the call had. Argument names come from the tool's schema or a fixed list. Any other name the client sent is recorded as unrecognized_key. Argument values are not sent.
  • Feature choices: the message action, such as archive or trash, whether an MCP email listing used a typed search query, whether a send or forward was scheduled, which kinds of body an email or draft had (plain text, HTML, both or neither) but not the body itself, whether an MCP attachment download returned a resource link, inline content or text from a PDF page range, whether an email preview included the send card that some MCP apps show with a Send button, and whether an email was sent from that card. When Fluxmail refuses a send from that card, it records which rule applied: the email was already sent, is already scheduled, changed after the preview, could not be found, or the card had expired. It also records when it refuses a send because that card sent the same email in the previous ten minutes. Attachment destination is recorded on failures too. When an app loads the send card, Fluxmail records the load with the app's name and version.
  • Stdio startup diagnostics: the phase that failed, ready, or setup_required when Fluxmail starts before setup, fixed reasons for recognized instance and session failures, and allowlisted filesystem error codes. Phases cover permission options, instance selection, configuration and database initialization, authentication, mailbox selection and transport startup. Events indicate whether the data directory came from the environment or the default and whether instance selection was explicit or automatic. They also record the connection's permission profile (read-only, read-write, full or custom) and whether the launch command chose no profile, so the default applies. A custom policy is reported as custom without its list of permissions.
  • Installation size: the plan and mailbox and member counts when the MCP server starts. Plans are Personal, Pro, Team, Business, Enterprise or other. Mailbox connection and removal events include provider type and mailbox totals, including totals by provider.
  • Mailbox setup diagnostics: whether OAuth used the public URL or a local callback, a built-in or custom provider application, and a direct callback, pasted callback or timeout. Events can indicate reauthorization, a fixed setup stage and failure category, request duration, HTTP status and a bounded category for the CLI request deadline.
  • MCP client metadata: the client's self-reported name and version, limited to 128 and 64 characters, a client category such as claude_code, codex or cursor, and the MCP protocol version. The name and version retain the text the client supplies.
  • MCP OAuth activity: registration, sign-in, approval or denial, token requests and revocation, with outcomes and OAuth error codes. Token events distinguish code exchange from refresh.

Data excluded from anonymous telemetry

Except for the MCP client's self-reported name and version, the stdio permission profile and the rejected argument names described above, anonymous telemetry excludes command arguments, email and mailbox data, identifiers, search text, file paths, credentials, configuration values and request payloads. It never sends provider responses, original error messages or stack traces. Person profiles and GeoIP lookup are disabled.

Excluded details include server names and addresses, folder names and labels, attachment names and content, recipients, message and delivery IDs, account IDs, page tokens, instance names, session tokens and file contents. Setup events exclude pasted callback URLs, exact timeout settings and the request ID returned in CLI errors and local server logs. MCP OAuth activity excludes client names and URLs, redirect addresses, codes, tokens, permissions, mailboxes and email addresses; the client name and version on MCP connection, tool and send card events are the separate exception above. Startup plan reporting excludes license keys and addresses.

Feedback you choose to send

When you explicitly ask an MCP client to submit feedback, send_feedback sends the message, category, product type, package version and any approved diagnostics and reply address to the Fluxmail Cloud feedback intake. This works even when anonymous telemetry is disabled. Review and confirm the message, diagnostics and exact optional reply address before submitting. A client must never choose or attach a connected mailbox address automatically. You can choose another address or send feedback without one. Exclude mail contents, attachments, raw tool arguments, logs, credentials, full connected-account lists and conversation history. When telemetry is enabled, the operation event records success or failure without the message, category, diagnostics, reply address or feedback ID.

Cloud stores reports and the chosen reply address privately for 90 days. They also go to feedback service providers and may be included in team notifications. Those copies have separate retention policies. The privacy policy lists the service providers. The reply address is labeled user-provided and unverified. It is not verified, automatically emailed, or used for authentication or as an analytics identity. See sending feedback for supported diagnostics, submission limits, data-transfer details and the administrator setting that disables submissions.

Last updated