Read PDF Out Loud for AI assistants

Connect Read PDF Out Loud to an AI assistant, and it can add PDFs to your library, read their clean text and summaries, and send you a link to listen in a natural voice with every word highlighted.

Connect your assistant

1. In your assistant, add Read PDF Out Loud from its connector directory, or add a custom connector (an MCP server) with the address https://readpdfoutloud.com/mcp.
2. Sign in with Google or with a link sent to your email, as you do on the site. New here? That creates your free account.
3. Check what the assistant asks to do, then press Allow.

Disconnect it at any time from Connected apps in the account menu on readpdfoutloud.com.

It reads PDFs the way the site does: running headers, page numbers, footnotes, citation numbers and the reference list are left out, two-column pages are read in order, and equations are written the way you'd say them. PDFs your assistant adds are in your library, ready to listen to on any device.

Things to ask

• “Add this paper to Read PDF Out Loud so I can listen on my commute: https://arxiv.org/abs/1706.03762”
• “What do the figures in the paper I just added show?”
• “Summarize section 3 of the report I added yesterday, with page numbers.”
• “Give me a link to listen to the results section.”
• “Pick up the paper I was listening to where I left off.”
• “How many pages do I have left this month?”

For developers

The connector is a remote MCP server. It holds no session, answers every request with plain JSON, and speaks both the current protocol and the handshake-based versions most clients still use.

Endpoints

Scopes

An app that asks for no scope gets all three. Calling a tool without its scope answers 403 with error="insufficient_scope" and the scope to ask for.

A request

curl https://readpdfoutloud.com/mcp \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "MCP-Protocol-Version: 2026-07-28" \
  -H "Mcp-Method: tools/call" \
  -H "Mcp-Name: get_account" \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/call",
       "params": {"name": "get_account", "arguments": {},
         "_meta": {"io.modelcontextprotocol/protocolVersion": "2026-07-28",
                   "io.modelcontextprotocol/clientCapabilities": {}}}}'

Tools

Every result is structured JSON matching the tool's output schema, repeated as text for clients that read only text. Page numbers are the PDF's own, starting at 1.

add_pdf

Adds a PDF to the library and starts reading it. It answers at once; a typical paper takes a minute or two, and get_document says when it's ready.

• Give exactly one of url, file or content_base64. Links must lead straight to a PDF; other web pages aren't read.
• Links are fetched from Read PDF Out Loud's servers: public https addresses only, up to 3 redirects and 30 seconds.
• A new PDF's pages count toward the month's allowance when it's added, as on the site. Adding a PDF that's already in the library returns it, and nothing is counted again (already_in_library).

{
  "document_id": "3f9c2a71b8d0",
  "title": null,
  "filename": "1706.03762.pdf",
  "pages": 15,
  "status": "queued",
  "text_ready": false,
  "summaries": {"done": 0, "total": 0},
  "words": null,
  "listening_minutes": null,
  "listen_url": "https://readpdfoutloud.com/#/doc/3f9c2a71b8d0/p/1",
  "resume_url": null,
  "last_page": null,
  "added_at": "2026-09-30T09:12:04Z",
  "error": null,
  "check_again_in_seconds": 10,
  "already_in_library": false
}

get_document

Takes a document_id and returns the fields above (without already_in_library):

list_documents

Lists the library, most recently used first. query keeps PDFs whose title or file name contains it; limit is 1 to 50 (10 by default). Each entry has document_id, title, filename, pages, status, text_ready, added_at and last_page; total counts every match.

get_text

Text stops at the end of the page before the limit, and next_pages says which pages to ask for next (null at the end). A single page longer than the limit is cut at a paragraph, with truncated true. Before the text is ready, it answers not_ready.

For example, page 4 of “Attention Is All You Need”, asked for with "pages": "4" and "format": "spoken" (text shortened):

{
  "document_id": "3f9c2a71b8d0",
  "title": "Attention Is All You Need",
  "format": "spoken",
  "pages_included": "4",
  "next_pages": null,
  "page_count": 15,
  "truncated": false,
  "summaries_complete": true,
  "characters": 2425,
  "text": "Figure 2: (left) Scaled Dot-Product Attention. … We compute the matrix of outputs as:\n\nAttention of Q, K and V, equals softmax of the fraction Q times K transpose, over square root of d sub k, V. …"
}

get_contents

Takes a document_id and returns entries, each with its title, level (1 is the top), page and a listen_url that opens the reader there. The contents come from the PDF's own outline when it has one, and from its headings otherwise.

get_summaries

Takes a document_id and optional pages, and returns items in reading order, each with its page, kind (figure, chart, table, code, algorithm or equation), label (such as “Figure 2”), caption and summary. complete is false while summaries are still being written.

delete_document

Takes a document_id and deletes the PDF with everything made from it. It can't be undone, and pages already counted this month aren't given back. Assistants should confirm with the user first.

get_account

Returns the account's email, plan, pages_used this month, pages_allowance and pages_left, and the date the allowance resets_on.

Limits

Errors

When a tool can't do what was asked, its result has isError set and carries {"error": {"code", "message", …}}: a sentence to show the user, and a code for the assistant.

If the connection has expired or been removed, requests are refused with 401 and the assistant asks the user to connect again.

Privacy

• An assistant you connect sees only your library, and only what you allowed.
• PDFs it adds are private to your account, like everything else in your library.
• We don't sell your data, and we don't use your documents to train AI models.
• Disconnect an assistant from Connected apps, or delete your account and everything in it, at any time.

The privacy policy and terms cover the connector.

Support and changes

Questions, or building on the connector? Write to support@readpdfoutloud.com.

• 1.0: the eight tools above.