soundlike

Docs

Soundlike for agents.

One MCP server, five tools, no accounts. The voice key returned by learn_voice is the whole credential.

Endpoints

MCP (Streamable HTTP, stateless): https://getsoundlike.com/api/mcp

OpenAPI mirror: https://getsoundlike.com/openapi.json

REST mirror: https://getsoundlike.com/api/v1/learn, /brief, /critique

Authentication

Bearer token. The token is the voice key from learn_voice, prefix sl1. Send it as Authorization: Bearer <voice key>, or pass it as the voice_key argument on any tool. Both work. No OAuth, no signup. tools/list needs no auth and returns the five tools.

The flow

  1. Call learn_voice once with the person's own writing. Best source: their last 20 to 30 posts from a connected account.
  2. Store the returned voice_key as the connector credential. Show them card_url; it's theirs to share.
  3. Before every draft, call get_voice_brief with the task and follow it exactly.
  4. Before anything is sent, call critique_draft. Use the rewrite when the score is under 85.

Tools

ToolArgumentsWhat it does
learn_voicename, samples[]Learns a voice from 3 to 60 verbatim samples. Returns voice_key, card_url, and the brief. Costs one model call, about 13 seconds.
get_voice_briefvoice_key?, task?, format?, length?, platform?The writing brief for the voice, optionally shaped to a task. Free, instant, deterministic. Call before every draft.
critique_draftvoice_key?, textScores a draft 0 to 100 against the voice, lists the spans that give it away with fixes, and returns a full rewrite. Call before anything is sent.
add_samplesvoice_key?, samples[]Relearns with more samples. Returns a new voice_key that replaces the old one.
voice_cardvoice_key?The shareable card link and image. Carries no credential.

Limits and plans

Data

Nothing is stored server side. The profile lives inside the signed key. Samples and drafts go to the model for one call and are not kept. Full detail in the privacy policy.

Questions: support@getsoundlike.com