renderkit-mcp

Model Context Protocol server for RenderKit. Gives AI agents — Claude Desktop, Claude Code, Cursor, and any MCP client — four tools:

| Tool | What it does | |---|---| | screenshot | Capture a page or HTML as a hosted PNG/JPEG/WebP. | | pdf | Render a page or HTML to a hosted, print-grade PDF. | | extract | Read a page as clean markdown / text / HTML (token-aware chunking optional). | | get_job | Poll an async render by id. |

It wraps the public RenderKit REST API (auth via x-api-key) — no wrapper code, no engine access.

Install / run

Runs straight from npm with npx; nothing to install globally.

// Claude Desktop / Claude Code / Cursor MCP config
{
  "mcpServers": {
    "renderkit": {
      "command": "npx",
      "args": ["-y", "renderkit-mcp"],
      "env": { "RENDERKIT_API_KEY": "rk_live_…" }
    }
  }
}

Configuration

| Env var | Default | Purpose | |---|---|---| | RENDERKIT_API_KEY | — (required) | Your RenderKit API key (rk_live_… / rk_sandbox_…). | | RENDERKIT_API_BASE | https://api.renderkit.tech | Override for staging / self-host. |

Example

Screenshot https://stripe.com full-page with ads blocked, then extract the pricing page to markdown and summarize the tiers.

The agent calls screenshot { url, full_page: true, block_ads: true } then extract { url, format: "markdown" } and reads the result.

Errors

Tool errors are returned as readable text including the RenderKit error code and request_id, so an agent can recover (retry, change params, or report).

Local dev

cd backend/mcp
npm install
RENDERKIT_API_KEY=rk_sandbox_… node src/index.js   # speaks MCP over stdio