My blog

Connect Claude to your blog

Your blog speaks MCP, the protocol Claude and ChatGPT use for tools. Connecting is one address and one approval.

The address

https://<your blog>/mcp

The app shows it with a copy button under Agents.

In claude.ai

  1. Settings โ†’ Connectors โ†’ Add custom connector.
  2. Give it a name (that's the name you'll see on its revisions) and paste the address.
  3. Claude opens your blog's consent page. Sign in if you aren't, pick which blog it may write for, and approve.

The connector then shows up in claude.ai, and Claude can use it in any chat where you turn it on.

In Claude Code

claude mcp add --transport http my-blog https://<your blog>/mcp

The first call opens the same consent page in your browser.

What the agent sees

The first thing a good agent does is call get_blog. It answers with the blog's name and address, the agent's level and what that level means, and how many posts are in each state. From there:

  • list_posts, read_post, list_revisions: reading.
  • create_draft, edit_post, submit_for_review: writing.
  • publish_post, schedule_post, unpublish_post, trash_post, restore_post: these run at once only at a high enough level. Below that they come back as waiting for owner with a request id, which lands in your inbox.
  • upload_media, confirm_media, upload_media_from_url, list_media: pictures.

Every change the agent makes needs the revision it read. If you changed the post in the meantime, the agent is told to read it again. Nothing gets silently overwritten.

A new connection starts at Propose

Propose means it writes drafts and suggests, and everything that reaches readers waits for you. Raise it under Agents if you want. Propose, Publish, Full explains each.

Taking it back

Agents โ†’ Revoke. The agent's next call is refused, and it can't refresh its way back in. Its posts and their history stay.