PDF in.
Clean Markdown out.

REST in one call

POST a PDF URL, get the canonical ExtractedDocument back — synchronously. Doc Links have REST endpoints too (/api/v1/links) — each tool below shows its REST mirror. OpenAPI spec (JSON)

curl -X POST https://doc.page/api/v1/extract \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer dk_live_..." \
  -d '{
    "url": "https://example.com/report.pdf",
    "outputs": ["markdown", "elements", "chunks"]
  }'

{
  "schemaVersion": "1.0",
  "source": { "engine": "server-pdfjs", "pages": 12 },
  "markdown": "# Title\n...",
  "elements": [{ "id": "el_0001", "type": "heading", "page": 1, "text": "Title" }],
  "chunks":   [{ "id": "ch_0001", "text": "...", "tokens": 486, "page": 1,
                 "section": "Introduction", "elementIds": ["el_0002"] }]
}

Native MCP server

Point your agent at the endpoint and it can read PDFs — and publish tracked Doc Links — by itself. Works with Claude Code, Claude Desktop, Cursor and any MCP client.

claude mcp add --transport http docpage https://doc.page/api/mcp

{
  "mcpServers": {
    "docpage": {
      "type": "http",
      "url": "https://doc.page/api/mcp",
      "headers": { "Authorization": "Bearer dk_live_..." }
    }
  }
}

What you get back

Every response is the same canonical object, whatever the engine:

• markdown

  the whole document as clean Markdown

• elements[]

  reading-order blocks with type and page (heading, paragraph, list…)

• chunks[]

  embedding-ready text pieces with token estimate, section and traceable element ids

• confidence

  the engine’s own honesty score — low table confidence means switch to hybrid mode

Two engines, one schema: the default "fast" engine is prose-focused (no tables/bbox); "hybrid" runs a heavier semantic engine that reconstructs tables and bounding boxes. If hybrid is temporarily unavailable, responses fall back to fast with an explicit warning.

Built for real-world documents

Limits & plans

Create and manage keys in your account settings ("API keys" tab). A page is counted per page of the processed PDF.

Errors

• 401 unknown or revoked key
• 429 rate limit reached — retry after the indicated seconds
• 402 monthly page quota used up