Why not just ask my model to write like someone?
You get an impression from memory: stock catchphrases, every format flattened into the same post, and your actual point lost somewhere in the costume. A Silvertongue voice hands your model the technique itself, studied from the source: how the person opens, argues and lands a line, and what they never do. Your facts, numbers and names come through exactly as you gave them.
How is a voice made?
From what the person actually published, studied in the medium the voice is for. Each profile breaks their writing down into stance, construction, rhythm, detail, formats and restraint, and links every source it was drawn from, so you can check the work yourself.
What can it write?
Posts, replies, explanations and threads, from a one-liner to a long piece, shaped for X, Threads, LinkedIn, Reddit, email or GitHub. Turn a voice down to a light touch, keep the edge clean for a work post, or switch to another voice mid-draft with one request.
Which agents does it work with?
Claude Code, Codex, VS Code and any other client that speaks MCP over Streamable HTTP connect to https://mcp.silvertongue.dev/mcp, and Claude Code and Codex can also install it as a plugin. Outside MCP, every voice is a VOICE.md file you can download from its page and hand to any model.
What does it cost?
Nothing. The server is free with no account or API key, and your own model writes the draft, so there is no new bill. The profiles are CC BY 4.0 and the code is Apache-2.0.
Can I ask for a voice?
Yes. Request one on GitHub and say what you would write with it. Know someone's writing cold? Add the voice yourself: CONTRIBUTING.md has the format.