You ship things alone, and your launch posts are the part you least want to babysit. The quickest way to see whether an assistant can help is to try it once, on something harmless. This is the shortest path: a few exact steps, then your first write as a rehearsal. The step-by-step tutorial covers the same ground with more detail.

Step 1: Open Blowhorn 1.61 or later, once

Blowhorn desktop 1.61.0 writes a blowhorn-mcp command to ~/.local/bin each time the app opens. It starts a local MCP server on your computer, so nothing needs hosting. Open the app, then check the file is there:

output 
ls -l ~/.local/bin/blowhorn-mcp

If it is missing, update the app first. Update Blowhorn shows how.

Step 2: Add it to your assistant

For Claude Code, one line:

output 
claude mcp add blowhorn -- ~/.local/bin/blowhorn-mcp

For Claude Code, Claude Desktop, Cursor or VS Code, the same command goes into that app's MCP settings. The connect guide has the exact entry for each, and for other clients. Restart the assistant if it asks you to.

A note on what is not available yet: running it with npx, a hosted server and catalog listings are coming soon. For now, you use the local command.

Step 3: Check it connected

In Claude Code, list your servers:

output 
claude mcp list

You should see blowhorn as connected. In other apps, look for it in the tools or MCP panel. It exposes 113 tools, one per Blowhorn command. If it shows as failed, open Blowhorn again and recheck the path. The MCP server reference lists the details.

Step 4: Make your first ask read-only

Begin with something that cannot change anything, and say so:

output 
Read only: list my Blowhorn profiles and what is queued for the next three days.
Do not change anything.

Reads run as asked. You are checking two things: the assistant can reach Blowhorn, and the profiles you see are the ones you expect. If a profile is missing, check the profiles reference first.

Step 5: Your first write, as a dry run

Now ask for a small post, and ask for the rehearsal in the same breath:

output 
Draft one short post saying I am trying Blowhorn with an assistant, for my
personal profile, scheduled for tomorrow at 10:00. Do a dry run, show me the
exact text, profile and time, and wait for me to say "go". Do not confirm yet.

Every write comes back as a dry run until the assistant runs it again, confirmed. What you see is the payload as it would be queued. Check the profile, the time and the text. Change anything you dislike by asking for a new dry run.

Step 6: Confirm

When it looks right:

output 
Go. Confirm it, then list what is queued and when it fires.

The post is now in your queue for tomorrow. The phrase "wait for me to say go" is what keeps the assistant honest; leave it out and you are relying on the default alone, which is safe but easier to misread.

If something goes wrong

  • Nothing is listed. Open the app and try the read again.
  • The post is wrong. Ask for a new dry run. Nothing was sent.
  • You want it all to stop. Pause stops everything.

When you are comfortable, read Why your assistant rehearses first for what each rehearsal protects, and Your release, drafted by your assistant, fired by you for a real launch.