Audiobooks cost real money. A single title runs $15 to $25 on the big platforms, and that internal report or 90-page PDF your friend sent you will never become one. Meanwhile, you’re already paying for an AI subscription that can do the whole job for about a dollar.
A new service called PDF TO MP3 hit Hacker News this week with a simple pitch: your AI agent converts documents to audio, your existing subscription covers the brainwork, and a cheap API key covers the voice. No new monthly plan. No editing software. I walked through the whole setup, and it’s genuinely beginner-friendly, so here’s how to turn a PDF into an audiobook without buying anything else.
What PDF TO MP3 actually is
It’s an MCP server, which sounds technical but isn’t. MCP stands for Model Context Protocol, a standard way for AI agents to plug into outside tools. Think of it like a USB port for your AI: once something is plugged in, the agent can use it when you ask.
So the flow looks like this. You attach a PDF in your AI chat. Your agent, which you already pay for through Codex or Claude, reads and processes the document. Then it calls the PDF TO MP3 tool, which converts the text to speech using a voice API key you provide. The service itself charges nothing. You pay your normal subscription plus pennies of speech generation.
Here’s the honest catch before we go further. Setup takes about ten minutes, and you’ll create an OpenAI API key if you don’t have one. API keys feel scary to beginners, but the whole thing costs less than a coffee per book, and you can set a spending cap so there are no surprises.
How to turn a PDF into an audiobook, step by step
Step 1: Connect the MCP server to your AI
In Codex, open Settings, then Plugins, then Add MCP server. Pick Streamable HTTP as the type and paste the server address from the PDF TO MP3 site. Save it, hit Authenticate, and connect your account. Claude users follow the same idea through the connector settings, and the site shows both paths with screenshots.
Step 2: Add a speech API key
This key pays for the voice only. Choose OpenAI or Gemini as your speech provider, grab an API key from their dashboard, and paste it into PDF TO MP3. The processing side stays on your subscription, which is why the service can stay free.
One tip here: OpenAI lets you set a monthly spending limit when you create the key. Set it to $5. That’s the entire worst-case scenario for a month of heavy converting.
Step 3: Attach a PDF and ask for your audiobook
Drop a PDF into your AI chat and type: “Use the PDF TO MP3 MCP to turn this into an audiobook.” The agent reads the document, breaks it into chunks, generates the narration, and hands you an MP3. A short PDF finishes in a few minutes. A full book takes longer and comes back split into chapter-sized files, which is actually what you want for a 10-hour listen.
What it costs to convert a PDF into an audiobook
OpenAI’s speech pricing is usage-based, and the current model, gpt-4o-mini-tts, works out to roughly $0.015 per minute of finished audio, per OpenAI’s own estimate. Longer documents use the same math: words become audio minutes, audio minutes become cents.
| Document | Approx. words | Audio length | Est. cost (gpt-4o-mini-tts) |
|---|---|---|---|
| 10-page report | 3,000 | 20 min | ~$0.30 |
| 50-page guide | 15,000 | 100 min | ~$1.50 |
| 300-page book | 90,000 | 10 hours | ~$9.00 |
Compare that with the Audible credit math or hiring a narrator, and the gap is absurd. Older OpenAI voices exist too: tts-1 runs $15 per million characters and the HD version $30, but for most listeners gpt-4o-mini-tts sounds better and costs about the same. It’s also steerable, meaning you can write instructions like “calm narrator, slightly slow, warm tone” and the voice follows along. If you want to check the current rates yourself, OpenAI’s text-to-speech guide has the full table.
Where this beats a normal TTS website
Plenty of sites convert PDFs to speech, so why bother with an agent? Three reasons, and the first one surprised me.
Cleanup comes free. Plain converters read everything: headers, page numbers, table junk, “click here” links. An agent can strip that noise first, fix broken sentences, and even summarize long footnotes before narrating. The audio sounds like it was prepared, not just dumped through a robot.
Then there’s the batching. Ask for five documents at once and the agent queues them while you do something else. And the files land where you’re already working. No upload portals, no watermarked free-tier MP3s, no account #14 to create.
Getting the narration quality right
Default voices are fine, but two small tweaks make a big difference. First, tell the agent your preference up front: “narrate with a friendly, medium-paced voice.” gpt-4o-mini-tts takes style instructions, so this actually changes the output. Second, feed it clean sources. A well-formatted PDF beats a scanned one, though agents handle messy docs better than any converter site I’ve tried.
If voice quality is your top priority, dedicated platforms like ElevenLabs still sound more human, and we covered how Gemini’s text-to-speech works for short clips. For commute listening, the cheap voices crossed the “totally fine” line a while ago.
Pro tips before your first real conversion
- Start with a 10-pager. Test the whole pipeline on something small, check the MP3, then commit to the 300-page monster.
- Set the spending cap on your API key before the first run. Five dollars covers a full book.
- Name your chapters. Ask the agent to split by chapter so your player shows clean files instead of one giant track.
- Listen at 1.2x. AI narration holds up better at higher speeds than you’d expect, and a 10-hour book becomes eight.
One more thing worth saying. This trick isn’t only for books. Meeting notes, course PDFs, that 40-page lease agreement you keep avoiding: attach, ask, listen on the drive home. Once the MCP is connected, converting any PDF into an audiobook is a one-sentence job.
The takeaway
You’re already paying for the smartest reader on the planet. Turning a PDF into an audiobook costs about $1.50 in voice credits and ten minutes of setup. Point your agent at the PDF pile, add the key, and send one short document today. Your commute is about to get a lot more productive, and if you’d rather hear characters than reports, see how Audible is adding AI characters to audiobooks for the other end of the spectrum.
