Ghost Lexical

Common filter patterns:

• All drafts: status:draft
• Published posts with a tag: tag:engineering+status:published
• Posts updated recently: use order: "updated_at desc" with a small limit

get_post

Retrieve a single post with full Lexical content. Accepts either a 24-character hex ID or a slug.

Parameters:

• id_or_slug (string, required) — Post ID (24-char hex like 6721a3b8e4f5c6d7e8f90a1b) or URL slug (like my-first-post)

Returns: Full post object including lexical field containing the JSON document as a string. Always parse the lexical field with JSON.parse() before reading or modifying content.

create_post

Create a new post. Defaults to draft status.

Parameters:

• title (string, required) — Post title
• lexical (string, optional) — Content as a JSON string of a Lexical document. Must be JSON.stringify()'d.
• status (enum: draft | published | scheduled, default draft) — Post status
• tags (string[], optional) — Tag names to assign. Tags are created automatically if they don't exist.
• custom_excerpt (string, optional) — A short summary/teaser for the post. Displayed in post lists, social cards, and SEO previews. Max 300 characters recommended.

Important: The lexical value must be a JSON string, not a raw object. Always wrap the Lexical document with JSON.stringify().

Large document strategy: If the Lexical JSON is very large (e.g., 30K+ characters for a long-form article), the MCP tool call may hit parameter size limits. In that case, use a two-step approach:

1. Call create_post with just the title, tags, status, and custom_excerpt (no lexical)
2. Call update_post with the post's id, updated_at, and the full lexical content

This separates metadata from content and avoids oversized single calls.

Escape hatch — direct MCP server invocation: If both create_post and update_post tool calls fail due to parameter size limits, you can invoke the ghost-mcp server directly via its stdio transport as a last resort. Write a small script that sends JSON-RPC messages to the server process:

// Run with: bun run create-large-post.ts
import jwt from "jsonwebtoken";

// Or use the MCP stdio protocol directly:
// 1. Spawn: bunx @perezd/ghost-mcp serve
// 2. Send JSON-RPC initialize handshake
// 3. Send notifications/initialized
// 4. Send tools/call with your create_post or update_post arguments
// 5. Read the JSON-RPC response from stdout

// Alternatively, call the Ghost Admin API directly:
const config = await Bun.file(
  `${process.env.HOME}/.ghost-mcp/config.json`
).json();
const site = config.sites[config.default];
const [id, secret] = site.apiKey.split(":");
const token = jwt.sign({}, Buffer.from(secret, "hex"), {
  keyid: id,
  algorithm: "HS256",
  expiresIn: "5m",
  audience: "/admin/",
});

const lexical = await Bun.file("my-large-post.json").text();

const res = await fetch(
  `${config.default}/ghost/api/admin/posts/`,
  {
    method: "POST",
    headers: {
      Authorization: `Ghost ${token}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      posts: [
        {
          title: "My Post",
          lexical: lexical,
          status: "draft",
          custom_excerpt: "My excerpt",
        },
      ],
    }),
  }
);
const data = await res.json();
console.log(data.posts[0].id);

This bypasses MCP tool parameter limits entirely by going straight to the Ghost Admin API. Use the same JWT signing approach that ghost-mcp uses internally. This should only be necessary for exceptionally large documents.

update_post

Modify an existing post. Requires updated_at for optimistic locking — Ghost rejects updates if the timestamp doesn't match, preventing overwrites of concurrent edits.

Parameters:

• id (string, required) — Post ID (24-char hex)
• updated_at (string, required) — The current updated_at value from the post. Obtain this from get_post or list_posts immediately before updating.
• title (string, optional) — New title
• lexical (string, optional) — New content as Lexical JSON string. Replaces the entire document — there is no partial/append mode.
• status (enum: draft | published | scheduled, optional) — New status
• tags (string[], optional) — New tag names. Replaces all existing tags.
• custom_excerpt (string, optional) — New excerpt text. Replaces the existing excerpt.

Collision detection workflow:

1. Call get_post to get the current post (and its updated_at)
2. Make your modifications to the Lexical content
3. Call update_post with the updated_at value from step 1
4. If Ghost returns a conflict error, re-fetch and retry

delete_post

Permanently delete a post.

Parameters:

• id (string, required) — Post ID (24-char hex)

Warning: This is irreversible. Always confirm with the user before deleting.

Lexical Document Structure

All Ghost content uses a Lexical JSON document format. The complete reference is in references/lexical-format-guide.md. Below are the essential patterns for constructing documents.

Root Structure

Every Lexical document has this wrapper:

{
  "root": {
    "children": [],
    "direction": "ltr",
    "format": "",
    "indent": 0,
    "type": "root",
    "version": 1
  }
}

All content nodes go inside root.children.

Node Types

Paragraphs

{
  "type": "paragraph",
  "children": [
    {
      "type": "extended-text",
      "text": "Your paragraph text here.",
      "format": 0,
      "detail": 0,
      "mode": "normal",
      "style": "",
      "version": 1
    }
  ],
  "direction": "ltr",
  "format": "",
  "indent": 0,
  "version": 1
}

A paragraph can contain multiple extended-text and link children to mix formatting and links inline.

Headings

{
  "type": "extended-heading",
  "tag": "h2",
  "children": [
    {
      "type": "extended-text",
      "text": "Section Title",
      "format": 0,
      "detail": 0,
      "mode": "normal",
      "style": "",
      "version": 1
    }
  ],
  "direction": "ltr",
  "format": "",
  "indent": 0,
  "version": 1
}

Tags: h1 through h6. Use h2 for main sections, h3 for subsections.

Text Formatting (Bitwise Flags)

The format field on extended-text nodes uses bitwise flags:

Combine with addition: bold + code = 18.

To mix formatting within a paragraph, use multiple extended-text children with different format values.

Links

Links are inline children within paragraphs:

{
  "type": "link",
  "url": "https://example.com",
  "children": [
    {
      "type": "extended-text",
      "text": "link text",
      "format": 0,
      "detail": 0,
      "mode": "normal",
      "style": "",
      "version": 1
    }
  ],
  "direction": "ltr",
  "format": "",
  "indent": 0,
  "version": 1,
  "rel": null,
  "target": null,
  "title": null
}

For bare URLs (no custom text), use "children": [].

Lists

{
  "type": "list",
  "listType": "bullet",
  "start": 1,
  "tag": "ul",
  "children": [
    {
      "type": "listitem",
      "value": 1,
      "children": [
        {
          "type": "extended-text",
          "text": "First item",
          "format": 0,
          "detail": 0,
          "mode": "normal",
          "style": "",
          "version": 1
        }
      ],
      "direction": "ltr",
      "format": "",
      "indent": 0,
      "version": 1
    }
  ],
  "direction": "ltr",
  "format": "",
  "indent": 0,
  "version": 1
}

For ordered lists: "listType": "number", "tag": "ol". List item value fields should be sequential (1, 2, 3...).

Images

{
  "type": "image",
  "version": 1,
  "src": "https://example.com/photo.jpg",
  "width": 1200,
  "height": 800,
  "title": "",
  "alt": "Descriptive alt text",
  "caption": "",
  "cardWidth": "regular",
  "href": ""
}

cardWidth options: regular, wide, full. Captions accept HTML.

Code Blocks

{
  "type": "codeblock",
  "version": 1,
  "code": "const greeting = 'hello';",
  "language": "javascript",
  "caption": ""
}

Use the appropriate language identifier for syntax highlighting.

Bookmark Cards

Rich link previews. Ghost auto-fetches metadata when you provide a URL:

{
  "type": "bookmark",
  "version": 1,
  "url": "https://example.com/article",
  "metadata": {
    "icon": "",
    "title": "Article Title",
    "description": "Article description",
    "author": "",
    "publisher": "",
    "thumbnail": ""
  },
  "caption": ""
}

When creating bookmarks, you can provide just the url and minimal metadata — Ghost will enrich it.

Signup Cards

Newsletter subscription forms:

{
  "type": "signup",
  "version": 1,
  "alignment": "left",
  "backgroundColor": "#F0F0F0",
  "backgroundImageSrc": "",
  "backgroundSize": "cover",
  "textColor": "#000000",
  "buttonColor": "accent",
  "buttonTextColor": "#FFFFFF",
  "buttonText": "Subscribe",
  "disclaimer": "<span style=\"white-space: pre-wrap;\">No spam. Unsubscribe anytime.</span>",
  "header": "<span style=\"white-space: pre-wrap;\">Sign up for updates</span>",
  "labels": [],
  "layout": "wide",
  "subheader": "",
  "successMessage": "Email sent! Check your inbox to complete your signup.",
  "swapped": false
}

Toggle / Accordion

Collapsible content sections. Both heading and content accept HTML:

{
  "type": "toggle",
  "version": 1,
  "heading": "<span style=\"white-space: pre-wrap;\">Click to expand</span>",
  "content": "<p>Hidden content here</p>"
}

Blockquotes

{
  "type": "extended-quote",
  "children": [
    {
      "type": "extended-text",
      "text": "Quoted text goes here.",
      "format": 0,
      "detail": 0,
      "mode": "normal",
      "style": "",
      "version": 1
    }
  ],
  "direction": "ltr",
  "format": "",
  "indent": 0,
  "version": 1
}

A container node like paragraphs — children can include extended-text and link nodes with mixed formatting.

Aside

A styled tangent or editorial aside:

{
  "type": "aside",
  "children": [
    {
      "type": "extended-text",
      "text": "An editorial tangent or side note.",
      "format": 0,
      "detail": 0,
      "mode": "normal",
      "style": "",
      "version": 1
    }
  ],
  "direction": "ltr",
  "format": "",
  "indent": 0,
  "version": 1
}

Same structure as a paragraph but rendered with distinct visual styling.

Callout Cards

Highlighted callout boxes with optional emoji and background color. The calloutText field accepts HTML:

{
  "type": "callout",
  "version": 1,
  "calloutText": "<p><strong>Note</strong> — This is important context.</p>",
  "calloutEmoji": "",
  "backgroundColor": "grey"
}

backgroundColor options include: grey, white, blue, green, yellow, red, pink, purple, and accent.

Embed Cards

oEmbed content (YouTube, Twitter, etc.). Ghost fetches the embed metadata from the URL:

{
  "type": "embed",
  "version": 1,
  "url": "https://www.youtube.com/watch?v=VIDEO_ID",
  "embedType": "video",
  "html": "<iframe ...></iframe>",
  "metadata": {
    "title": "Video Title",
    "author_name": "Channel Name",
    "author_url": "https://www.youtube.com/@channel",
    "type": "video",
    "height": 113,
    "width": 200,
    "version": "1.0",
    "provider_name": "YouTube",
    "provider_url": "https://www.youtube.com/",
    "thumbnail_height": 360,
    "thumbnail_width": 480,
    "thumbnail_url": "https://i.ytimg.com/vi/VIDEO_ID/hqdefault.jpg",
    "html": "<iframe ...></iframe>"
  },
  "caption": ""
}

When creating embeds, provide the url and embedType at minimum — Ghost enriches the metadata.

HTML Cards

Raw HTML injection with optional visibility controls for member segmentation:

{
  "type": "html",
  "version": 1,
  "html": "<div>Custom HTML content here</div>",
  "visibility": {
    "web": {
      "nonMember": true,
      "memberSegment": "status:free,status:-free"
    },
    "email": {
      "memberSegment": "status:free,status:-free"
    }
  }
}

The visibility field is optional. When present, it controls who sees the content:

• web.nonMember — whether non-members see this block on the website
• web.memberSegment / email.memberSegment — NQL segment filter (e.g., status:free for free members, status:-free for paid members)

If visibility is omitted, the HTML block is shown to everyone.

Line Breaks (Horizontal Rules)

{
  "type": "linebreak",
  "version": 1
}

Required Fields Checklist

Every container node (paragraph, heading, list, listitem, extended-quote, aside) requires all of these:

• type — node type identifier
• version — always 1
• children — array of child nodes
• direction — "ltr" for left-to-right
• format — "" (empty string for container nodes)
• indent — 0 for no indentation

Every extended-text node requires:

• type — "extended-text"
• text — the text content
• format — bitwise format flag (0 for plain)
• detail — 0
• mode — "normal"
• style — ""
• version — 1

Card nodes (image, codeblock, bookmark, signup, toggle, callout, embed, html, linebreak) are simpler — they require type, version, and their specific fields.

Content Construction Patterns

Building a Complete Post

1. Start with the root wrapper
2. Add nodes to root.children in reading order
3. Stringify the entire document: JSON.stringify(doc)
4. Pass the string to create_post or update_post as the lexical parameter

Editing an Existing Post

1. get_post to retrieve current content
2. Parse the lexical field: JSON.parse(post.lexical)
3. Navigate root.children to find and modify nodes
4. Re-stringify and call update_post with the post's current updated_at

Typical Blog Post Structure

A well-formed blog post typically follows this node sequence in root.children:

1. Opening paragraph(s) — hook and context
2. extended-heading (h2) — first section
3. Paragraphs with mixed formatting and links
4. extended-quote for blockquotes, aside for editorial tangents
5. list nodes for enumerations
6. image, embed, or codeblock nodes as needed
7. callout cards for highlighted notes or warnings
8. linebreak between major sections
9. More h2/h3 sections as needed
10. Closing paragraph(s)
11. Optional html card for author bio or custom blocks (with visibility controls)
12. Optional signup card at the end

NQL Filter Reference

The list_posts tool supports Ghost's NQL (Neon Query Language) for filtering:

• Field match: status:draft, tag:news
• Combining filters: tag:news+status:published (AND), tag:news,tag:tech (OR)
• Negation: -status:draft (NOT draft)
• Comparison: created_at:>'2025-01-01'
• Author filter: author:derek

Common queries:

• All drafts: status:draft
• Published with specific tag: tag:my-tag+status:published
• Recent posts: use order: "published_at desc" with limit: 5