Agent access (MCP server)
Pro and Team include an MCP server, so an AI agent such as Claude Code can work with your publishing calendar: read your content, schedules, and approvals and, if you allow it, schedule posts or approve what is waiting. You decide what each agent may do, and newsletters always stay with you.
Set it up
- In Postwright, open Settings, then Agent access. Name the token (for example “Claude Code on my laptop”), choose its scopes, and create it.
- Copy the token, or the ready-made command under it, now. Postwright keeps only a hash and can’t show it again. If you lose it, revoke it and make another.
- Run the command in a terminal. It is a
claude mcp add postwrightline that uses the app’s own bundled Node, so you don’t need Node installed. The card shows the exact paths for your install. - Start Claude Code and ask “What’s scheduled on Postwright next week?” Running
/mcpshows the server and its tools.
Other MCP clients can run the same command over stdio with the same environment variables.
Scopes
| Scope | Default | What the agent can do |
|---|---|---|
read | Always on | Read content and its publications, schedules, the review queue, the calendar, the cadence settings, pending newsletter sends, and the posts drafted for you to post by hand (Substack Notes). It can remind you about one but never record it as posted. |
schedule | Off | Schedule approved content on one of its channels, move a schedule that hasn’t run yet, or cancel one. Without approve, a new schedule is only a proposal that waits in your review queue until you approve it. |
approve | Off | Approve or reject schedules waiting for review (approving publishes the post at its time), and cancel a pending newsletter send. The card warns you when you tick it. |
What an agent can never do
- Send a newsletter, or approve a newsletter schedule, with any scope. A send goes to your whole list and can’t be taken back, so it stays with you in Postwright.
- See or change connections, credentials, settings, your brands, or the license, or create, list, or revoke agent tokens.
- Publish directly, edit content, change the cadence, or bulk-approve.
- Read the comment Inbox, or queue or approve a reply to a comment.
- Schedule or touch the video pipeline and phone shorts, which are reviewed in Postwright itself.
Anything an agent reads is scrubbed of the secrets Postwright knows about before it is sent back, though it can’t catch a secret it doesn’t know.
Tokens stay on your Mac
Tokens are stored hashed, shown once, and work only from the computer Postwright runs on, reads included. A request that arrives through a tunnel that says so, or from another machine, is refused. One limit to know: Postwright can only tell a request is tunnelled when the tunnel says so, so while any token exists, don’t expose Postwright’s port through a tunnel that adds no forwarding headers (for example ssh -R or frp). Cloudflare Tunnel adds them. To change a token’s scopes, make a new token and revoke the old one.