A new tool called PDF TO MP3 lets people turn PDF documents into spoken audiobooks by working through an AI assistant they already pay for. According to the project's own page, the service itself charges nothing extra. Document processing runs through a user's existing Codex or Claude subscription, while speech generation is billed directly by OpenAI or Google based on usage.
The approach is notable less for the conversion itself than for how it is wired together. Instead of uploading files to yet another subscription service, users attach a PDF in their AI chat and ask the assistant to convert it. The project's page describes the whole process as triggered from that single chat request.
How the setup works
PDF TO MP3 connects to a user's Codex or Claude account through an MCP server, which the user configures in settings. MCP, or Model Context Protocol, is a standard that lets AI assistants communicate with outside tools. In this case it acts as the bridge between the chat interface and the audiobook converter.
Once the connection is in place, the assistant handles the document side of the job. Speech is produced separately, through the user's own OpenAI or Gemini API key. That split means the person using the tool deals with two accounts: the AI subscription that processes the text and the API account that generates the audio.
Who pays for what
The cost model is the feature most readers will focus on. The service does not add a fee of its own. Costs flow to the subscriptions people already hold, and to the API accounts they link for audio.
- AI backend: Codex or Claude, covered by an existing subscription.
- Speech generation: billed per usage by OpenAI or Google, depending on which API key is used.
- PDF TO MP3 service: no extra charge, according to the project page.
For anyone who listens to long documents, the practical question is how much audio generation will cost in total. The source does not give price figures, so people planning heavy use should check the current rates published by OpenAI or Google before starting.
A note on Gemini free-tier keys
The project warns that Gemini free-tier keys may run into rate limits. To keep conversions reliable, the page recommends enabling billing on the account. That advice matters for anyone who expects to convert a full-length book rather than a few pages, because a stalled job is more frustrating than a modestly higher bill.
Why the model is getting attention
Text-to-speech tools have been around for years, and many people use them to listen to articles, reports or study material. What is different here is the reliance on tools people already have. A reader who pays for an AI assistant may not want another monthly fee for a narrow task, and the project's pitch is that the work fits inside what they already buy.
The idea also reflects a wider pattern in software, where small utilities act as connectors between AI assistants and everyday tasks. Converting a document to audio is a simple example, but it shows how a single chat window can start to coordinate several services at once. Whether that becomes a common habit will depend on how easy such setups are for ordinary users.
The project drew attention on Hacker News, where the post had 125 points and 61 comments at the time the source was captured. Community discussion of this kind often raises practical questions about setup, cost and reliability that a product page does not fully answer.
Points to check before trying it
Readers considering the setup should weigh a few things. The workflow depends on an active Codex or Claude account, a correctly configured MCP server, and an API key from OpenAI or Google. Each of those is a separate component, so troubleshooting may involve more than one provider. Usage costs are variable and depend on how much text is converted.
Licensing is also worth considering. Users should convert only documents they have the right to use, a point that applies to any document-to-audio tool, not just this one.
The source material is a product page rather than an independent review, so its claims about cost and performance have not been tested by a third party. Whether the model holds up as more people try it will depend on how clearly providers price speech generation and how reliably MCP-based connections work in everyday use.
Related: Pac-Man Becomes a Real-Time Test for AI Decision Models · Free Tool Converts OpenAI, Flowise and Pipedream Workflows to n8n
Comments (0)