Driving your article audio from an AI agent (MCP)
The MCP protocol lets an agent like Claude voice an article, check a status or fetch subtitles in plain language. What that actually changes for a newsroom.
Most audio integrations are still imagined as code: an API, a CMS, a developer wiring one to the other. That works, but it assumes an available engineering team and a planned project. A different path has recently opened, and it looks nothing like a classic integration: driving audio generation straight from an AI agent, in plain language, through the MCP protocol.
MCP (Model Context Protocol) is an open standard that lets an assistant like Claude connect to an outside service and use its functions. Instead of writing a request, you describe what you want and the agent picks the right tool. WeDispatch exposes an MCP server at https://wedispatch.fr/api/mcp. Once the connector is added, eight tools become available in the conversation, and you never call any of them by hand.
What you can ask, in one sentence
The idea is to replace manipulation with intent. "Voice this article in the Pauline voice, Sports section" is enough to generate the audio, with its URL and its embed link. "Where is article 9f3c?" returns the status (processing, ready, stale or failed), the duration, the voice and the subtitles. "Regenerate this article in English" starts a new version while the old one keeps being served. "Which female French voices do I have?" lists your voice catalogue. "Which articles are most listened to this month?" returns plays, completion rate and registration-wall conversion.
These are not commands in disguise: they are ordinary requests that the agent turns into calls. The drop in friction is real. A desk editor who has never opened a piece of technical documentation can voice an article, check its status and grab a subtitle file without leaving the conversation.
Who this is really for
Let us be honest about the use cases. For a newsroom publishing dozens of articles a day, automation through an RSS feed remains the right answer: new articles are detected and voiced with nobody intervening. MCP does not replace that feed and is not meant to carry your volume for you.
Where it earns its place is everything that falls outside the automatic flow: reviving an older article that a news spike pushes back into view, testing a different voice on a particular format, producing the English version of a feature on demand, or simply checking listening statistics without opening the dashboard. It is a conversational control room, not a production line. For a reporter who wants to act fast on one specific article, it is often quicker than any interface.
Connection: one click, no key to copy
The technical barrier was deliberately lowered. You add an MCP connector pointing at the server URL, and on connection the agent redirects you to the usual WeDispatch sign-in screen (OAuth 2.1 with PKCE). You authenticate in your customer account, you authorise access, and that is it: no key to copy and paste, and access stays revocable at any moment. For clients that do not yet handle OAuth, API-key authentication is still available through the mcp-remote bridge.
That absence of a plain-text key matters for a newsroom: it keeps a token from sitting in a shared configuration file, and it lets each person act with their own permissions.
What it does not do
An honest article should state its limits. MCP does not create content for you and does not judge whether a voice is good: it drives generation, it does not invent it. It does not remove the need for an account or a character quota. And it is not a substitute for embedding the player on your site: that still goes through the script you paste or the embed link the tool returns. MCP acts on producing the audio and reading the statistics, not on the final display in front of your readers.
Put differently, it is a new front door, not a new product. The same generation, the same voices, the same subtitles as through the API or the RSS feed, but driven by conversation rather than by code.
Where to start
If you already use an MCP-compatible assistant, the trial costs a few minutes: add the connector, sign in, and ask it to voice a test article. You will see straight away whether this way of working suits you. The connection details and the full list of tools are documented on the API reference page, and automatic generation stays available in parallel through the automatic podcast for anything that is about volume.
Give your articles a voice with WeDispatch
The free tier lets you listen to your own articles today, and the plans, from 69 to 419 EUR excluding VAT per month, are detailed on the pricing page. To try agent-driven control, start here.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo