@pagehub/mcp-core

The toolset that powers PageHub's AI integrations — 98 typed tools for building, editing, and publishing websites programmatically.

This is the shared core: tool schemas (MCP-compatible), HTTP handlers against the PageHub API, request context, and validation helpers. It's consumed by:

• @pagehub/mcp — stdio MCP server (use with Claude Desktop, Cursor, etc.)
• PageHub's in-app AI agent (chat-driven editor)
• Your own Claude Agent SDK / OpenAI / Vercel AI scripts

What's PageHub?

A visual + AI website builder. The MCP tools let an LLM create sites, edit their HTML + Liquid files, place blocks, set themes, manage pages, run accessibility/SEO audits, search stock media, and publish — all against pagehub.dev via authenticated HTTP.

Install

npm install @pagehub/mcp-core
# or
pnpm add @pagehub/mcp-core

Node.js 18+.

Quick start

const { runWithContext, executeTool, getAllTools } = require("@pagehub/mcp-core");

// 1. Inspect available tools (MCP schema)
const tools = getAllTools(); // 98 tools

// 2. Run a tool inside a request context
const result = await runWithContext(
  {
    apiKey: process.env.PAGEHUB_API_KEY,
    apiBaseUrl: "https://pagehub.dev",
  },
  () => executeTool("list_sites", {})
);

console.log(result.content[0].text);

Get a free API key by signing up at pagehub.dev or calling the register tool with an email.

See examples/ for runnable scripts (list tools, execute a tool, full Claude Agent SDK loop).

Use with Claude Agent SDK

const { generateText, tool } = require("ai");
const { gateway } = require("ai");
const { getAgentTools, runWithContext, executeAgentTool } = require("@pagehub/mcp-core");

const schemas = getAgentTools(); // 67 tools safe for public agents

const aiTools = Object.fromEntries(
  schemas.map(s => [
    s.name,
    tool({
      description: s.description,
      parameters: s.input_schema,
      execute: args => executeAgentTool(s.name, args),
    }),
  ])
);

await runWithContext({ apiKey: process.env.PAGEHUB_API_KEY }, () =>
  generateText({
    model: gateway("anthropic/claude-sonnet-4-6"),
    tools: aiTools,
    prompt: "Create a landing page for a coffee shop.",
  })
);

Tool catalog

About 100 tools across 11 domains. The full schemas live in src/data/tools.json.

Domain	Examples
Sites	list_sites, create_site, select_site, publish_site, unpublish_site
Templates	list_templates, pull_template, save_template, update_template
Pages	list_pages, add_page, update_page, delete_page
Files	list_site_files, read_site_file, edit_site_file, edit_site_files
Blocks	search_blocks, get_block, save_block, patch_block, apply_kit_block
Theme & Design	set_theme, suggest_palettes, suggest_font_pairings, set_favicon
Stripe	stripe_connect, set_checkout, refresh_products, stripe_search_products
Media	upload_image, find_image, find_video (Pexels / Unsplash)
SEO	audit_seo, audit_accessibility
Portal	set_portal, get_portal, remove_portal
Site Config	set_integrations, set_redirects

Public API

const {
  // Tool execution
  runWithContext, // (ctx, fn) — seed request context (AsyncLocalStorage)
  getContext, // () — read current context
  executeTool, // (name, args) — execute any HTTP tool
  executeAgentTool, // (name, args) — execute only agent-allowed tools
  handlers, // raw handler map

  // Schemas
  getAllTools, // every tool, MCP format ({inputSchema})
  getAgentTools, // the public-agent-safe tools, Claude API format ({input_schema})

  // Sets
  HTTP_TOOL_NAMES,
  AGENT_ALLOWED,
  AGENT_EXCLUDED,

  // Helpers
  apiFetch,
  parseMaybeJson,
} = require("@pagehub/mcp-core");

Subpath imports

For handler delegation patterns (e.g. wrapping individual tools), import handlers directly:

const site = require("@pagehub/mcp-core/handlers/site-config"); // { set_integrations, set_favicon, set_redirects }
const seo = require("@pagehub/mcp-core/handlers/seo"); // { audit_seo, audit_accessibility }
const ai = require("@pagehub/mcp-core/ai-models"); // ESM — AI model defaults

Available subpaths: handlers/{discovery,portal,remote,remote-collections,seo,site-config,stock-images,stock-videos}, ai-models. Internals (core/, helpers/, utils/, validation/, data/, lib/) are not part of the public API.

Source layout

src/
├── index.js              ← public entry
├── ai-models.mjs         ← ESM: AI model defaults
├── core/                 ← context, api-fetch, direct upload
├── handlers/             ← HTTP tool implementations (one file per domain)
│   └── seo/              ← SEO + a11y audit (server-rendered draft page)
├── helpers/              ← shared handler utilities (target, file edits, library, publish state)
├── validation/           ← button, branding validators
├── utils/                ← pure utilities (color)
├── data/                 ← static data: tools.json, instructions, style reference, vibes, categories
└── lib/                  ← theme-fonts

See src/README.md for a per-folder breakdown.

Contributing

See CONTRIBUTING.md for the handler contract, context system, and how to add a new tool.

License

MIT