Connect an assistant and send your first post

Add the Blowhorn MCP server to Claude Desktop, read your profiles and queue, and queue a post that waits for your approval.

In this tutorial you connect Claude Desktop to Blowhorn over MCP. You ask it to read your profiles and your queue. Then you ask it to draft a post and queue it. You see the dry run first, you confirm it, and the post lands in your queue waiting for your approval. Nothing is published. It takes about ten minutes.

Before you start

You need:

• A Mac with the Blowhorn app, version 1.61.0 or later, installed and signed in to your Blowhorn account. See Install the app.
• At least one profile signed in to LinkedIn, and mapped to its Google Chrome profile. See Sign a profile in to each platform.
• Claude Desktop, installed and signed in.

1. Check the blowhorn-mcp command

1. Open the Blowhorn app once, then quit it if you like.

2. Open Terminal.

3. Run:

   $ ~/.local/bin/blowhorn-mcp --version

You see the Blowhorn version number. If you see No such file or directory, update the app to 1.61.0 or later and open it again.

2. Add Blowhorn to Claude Desktop

1. In Claude Desktop, open Settings > Developer > Edit Config.

2. Replace the file's contents with this entry. Put your macOS user name in place of <you>:

   {
     "mcpServers": {
       "blowhorn": {
         "command": "/Users/<you>/.local/bin/blowhorn-mcp",
         "args": []
       }
     }
   }

   If the file already lists other servers, add the blowhorn entry beside them inside mcpServers instead.

3. Save the file.

4. Quit Claude Desktop fully, then open it again.

3. Check that it connected

Start a new chat and ask:

Which Blowhorn tools do you have?

Claude lists tools with names such as profile_list, content_list and content_upsert. There is one tool for each Blowhorn command, except a few that uninstall Blowhorn or sign you out, which stay hidden unless you turn them on.

If Claude has no Blowhorn tools, see Troubleshooting in Connect an AI assistant over MCP.

4. Read your profiles and your queue

Ask:

List my Blowhorn profiles, then show me what is in my content queue.

When Claude asks to use a tool, allow it. Claude calls profile_list and then content_list. Both only read, so they run as asked. You see your profiles by name, and each activity in the queue with its state.

Pick one profile name from the list. This tutorial calls it <profile>.

5. Ask for a draft and queue it

Ask, with your profile name in place of <profile>:

Draft a short LinkedIn post for profile <profile> saying we are trying out Blowhorn. Queue it with content_upsert, but leave Approved? empty. Show me the dry run first.

Claude calls content_upsert. A tool that changes anything runs as a dry run unless the call confirms it, so nothing is written yet. Claude tells you it was a dry run and shows the post. The tool's own result looks like this:

{
  "activity": null,
  "dry_run": true,
  "expected_fingerprint": null,
  "ok": true,
  "payload": {
    "Message Text": "...",
    "Platform": "linkedin",
    "Profile": "<profile>"
  }
}

6. Review the dry run

Read the dry run. Check three things:

• Platform is linkedin.
• Profile is your profile.
• Message Text is the post you want.

There is no Approved? value, so the post will wait for you. If anything is wrong, ask Claude to change it and show the dry run again.

7. Confirm it

Ask:

That looks right. Run it again with confirm.

Claude calls content_upsert again with confirm set to true. This time the activity is written. Claude reports the new activity's number and its state, draft. In the tool's result, "created" is true and "state" is "draft".

draft means the activity is not approved, so no run picks it up.

8. See it in your queue

Ask:

Show me my content queue again.

The new activity is in the list, in state draft. Open the Blowhorn app's Content screen and you see it there too, waiting for approval.

What you did

You connected an assistant to Blowhorn, read your profiles and queue, and queued a post. The assistant saw a dry run before anything was written, and only your confirm wrote it. The post is still waiting for you. The assistant cannot approve it: the content_approve tool refuses an assistant in every approval mode.

When you want the post to go out, approve it yourself in the app, as in Approve a waiting activity. If you would rather not keep it, delete it from a terminal, with your activity number in place of <number>:

$ blowhorn content delete <number> --yes

Next steps

• Connect an AI assistant over MCP - set up ChatGPT, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Zed, Cline or Goose, and more example prompts
• How the MCP server keeps you in control - why dry runs, confirms and approvals work the way they do
• MCP server reference - every tool and option