Silvertongue MCP¶
An open catalogue of writing styles, versioned and sourced, that an agent can adopt for one post and drop for the next - with every voice's research linked rather than implied.
Silvertongue runs at https://mcp.silvertongue.dev/mcp over Streamable HTTP, keyless and free. A bare initialize answered HTTP 200 as silvertongue version 0.1.2 with tools, resources and prompts capabilities. resources/list returned 27 VOICE.md documents of about 3 to 4 KB each, and tools/list returned four tools.
Server type: Remote (Streamable HTTP)
Auth: none (free, read-only)
Endpoint: https://mcp.silvertongue.dev/mcp
Tools: 4
Resources: 27 VOICE.md profiles
Prompt: write_as
Registry identity: dev.silvertongue/silvertongue@0.1.0
Built by: Silvertongue (auscaster.com); repo github.com/auscaster/silvertongue
Why This Matters for Operators¶
"Write it in my voice" is the request every content workflow starts with and almost none of them satisfy, because the input is a sample paragraph and the output is a generic impression of energy. The root cause is that style is a set of decisions - sentence length, what gets cut, how a claim is introduced, when a number replaces an adjective - and a sample does not state which decision it is showing.
This server turns style into a studied, citable artifact: each voice documents what was read, what the observations were, a handwritten example, its strengths and its limits. That makes the output reviewable. When a draft reads wrong, the argument is about a rule in a document rather than about taste.
The second decision is provenance. Every response carries a release_id digest, so a voice pinned in one campaign can be reproduced after the catalogue moves. get_changes compares the IDs an agent has cached against the current release and reports which voices were added, re-studied as a new edition, or removed.
Tools and Capabilities¶
| Tool | What it does |
|---|---|
search_voices |
Find a voice by name, by lane, or by what the writing should do - a roast, a launch post, a plain explanation, a speech |
get_voice |
The full profile: VOICE.md, every source studied with a note on what it shows, the observations drawn, a handwritten example, licence and attribution |
get_voice_packet |
Instructions for writing in one voice at a format, length, platform, intensity and edge, carrying the full VOICE.md and an exact length target fitted to the platform's limit |
get_changes |
Diff cached voice IDs against the current release: added, re-studied, removed |
Lanes are tech, celebrity, comedy, classics and politicians. Each VOICE.md is also exposed as a resource, so a client can read a profile without a tool call. The write_as prompt switches the conversation to one voice in a single step.
The packet is the piece that makes this practical for operators. It carries the rules that keep the user's own facts and identity intact, and it opens with a line that replaces any earlier voice, so a post can be written in one style and the next in another without the conversation dragging the first one along.
What to Watch¶
The server states its own limit plainly: profiles are unofficial studies of public writing technique, and the people named did not write or endorse them. That is the correct framing and it should stay the framing in any published use. A marketing team adopting a named-public-figure voice for outbound content is making a brand and legal decision, not a stylistic one, and a voice that imitates a living person's public persona carries risks that a plain instruction like "write plainly, cut the adjectives" does not.
Licensing splits by asset: the code is Apache-2.0, the voice profiles are CC BY 4.0, which requires attribution when a profile is reused.
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.
See Also¶
FAQ¶
What is Silvertongue MCP?¶
A hosted, keyless, read-only MCP server serving an open catalogue of studied writing voices, so an agent can find a voice, read how it was researched, and write a draft in it with the user's own model.
Does it need an API key or an account?¶
No. The endpoint answered initialize with HTTP 200 without credentials, and all four tools are read-only.
How many voices does it carry?¶
Twenty-seven VOICE.md profiles were listed as resources at the time of this pass, across the tech, celebrity, comedy, classics and politicians lanes, plus a write_as prompt and a packet builder.
Can the output stay reproducible after the catalogue changes?¶
Yes. Every response carries a release_id digest that pins the catalogue release, and get_changes reports additions, new editions and removals against the IDs an agent has cached.
Can I use a named person's voice in my marketing?¶
The profiles are unofficial studies of public writing technique and the people named did not write or endorse them. Reusing a profile requires CC BY 4.0 attribution, and imitating a living person's persona for outbound content is a brand and legal decision rather than a formatting choice.