Skip to content

PRIVACY.md

Latest commit

 

History

History
87 lines (73 loc) · 5.95 KB

PRIVACY.md

File metadata and controls

87 lines (73 loc) · 5.95 KB
Raw

Privacy

substack-mcp is software you run yourself. The maintainer operates no hosted service and receives none of your data. The package contains no telemetry, analytics or crash reporting.

What the server and CLI send, and where

Destination When What is sent
The Substack publication you configure MCP tool calls, and CLI commands that read or write Substack data (status, --help and other local commands make no request) Your Substack session cookie and the request needed for that operation: reading posts, drafts, statistics or subscribers; creating or updating drafts; uploading images; publishing Notes; adding consented free subscribers
The public page of your publication, and any host it redirects to The subscriber-count read, when Substack's API does not return a count A plain HTTPS request with no cookie or credential. Up to 3 HTTPS redirects are followed, including to other hosts
substack.com, your publication, and whatever those pages load substack-mcp login only You sign in through a local browser window, which behaves like any browser: it follows the pages' redirects and loads the resources they request, including third-party hosts. The resulting session cookie for your publication is saved locally
The image host you name, and any host it redirects to upload_image with image_url only A plain HTTPS download request, with no Substack cookie or credential. Up to 3 HTTPS redirects are followed, including to other hosts; private and reserved network addresses are refused. The image is then uploaded to Substack's CDN, where anyone with the returned link can fetch it

With the core server and CLI, those are the only destinations. The optional components below contact other services when you set them up.

Your MCP client or AI assistant is a separate party. Content you give it (for example draft text) and every tool result the server returns to it (including private drafts, posts, statistics, subscriber records and error messages) are handled under that client's privacy terms. The CLI prints results to your terminal or to files you redirect it to. This server only sends your content to Substack when a tool you invoke requires it. Substack's handling of your data is governed by Substack's terms and privacy policy.

What is stored on your machine

  • Sessions and profiles. substack-mcp login saves your publication URL, user ID, session cookie and the save time in ~/.substack-mcp, or in the directory named by SUBSTACK_MCP_HOME. The file is encrypted with a key derived from your OS account and machine; see SECURITY.md for its limits. Saving a named profile first writes the encrypted file to a temporary .profile-<id>.tmp in the same directory and then removes it; if that cleanup fails (the CLI reports it), the encrypted temporary copy can remain. Credentials supplied through environment variables are read at startup and are not written to disk.
  • Files you ask for. export writes the output path you specify. Markdown export also writes <path>.source.json beside it, containing the complete export bundle including the original draft body. Each file is written through a temporary file in the same directory; if cleanup fails, the CLI reports it and the temporary copy can remain.
  • Logs and errors. Session tokens and cookies are not printed. Error output can include messages returned by Substack or by the network layer (for example when authentication fails at startup or an export fails), so review error output before sharing it.

Optional components you deploy yourself

These are off unless you set them up:

  • Local calendar sync (calendar-sync) runs your own gws command-line tool to read Google Calendar booking emails from Gmail, then adds consenting bookers as free Substack subscribers. Gmail is contacted through gws under your Google authorization. Its state file, which you place, contains the publication and organizer, bookers' email addresses, signup answers, source message IDs, and a record of each subscriber attempt with its status and time. Each save is written first to <state_path>.new; if writing or renaming fails, that copy can remain. See docs/calendar-sync.md.
  • Cloud calendar sync runs the same process in a Cloudflare Worker on your own Cloudflare account. It sends your Google OAuth client credentials and refresh token to Google's token endpoint (oauth2.googleapis.com), reads booking emails and can send report emails through the Gmail API (gmail.googleapis.com), and calls your Substack publication. Its Durable Object storage holds the consent ledger, run history, enabled and initialization state, and report delivery records (attempt times, status and the Gmail message ID of sent reports). Your credentials are kept as Cloudflare secrets. See docs/cloud-calendar-sync.md.
  • HTTP transport (MCP_TRANSPORT=http) listens on the address you choose and acts with your Substack session for requests it accepts. By default it accepts only loopback Host and Origin values; set MCP_HTTP_TOKEN to require a bearer token, and do not expose the port to untrusted networks. See docs/http-transport.md.

Your control

  • Revoke access by signing out of Substack, which invalidates the session cookie, and delete the session directory: ~/.substack-mcp, or your SUBSTACK_MCP_HOME directory if you set one.
  • Uninstall the package to stop all activity. There is no background service unless you deployed one of the optional components above; revoke their Google authorization and Cloudflare secrets separately.
  • Images uploaded to Substack's CDN and Notes you publish are public, and this server has no delete tools for them.

Contact

Report privacy or security concerns through GitHub's private vulnerability reporting on this repository (Security tab), as described in SECURITY.md.