Skip to content
KugelAudioDocs

MCP server

Connect Claude, Cursor or any MCP client to your KugelAudio account through the MCP server to generate speech from a chat, plus billing and access details.

On this page

The KugelAudio MCP server lets an AI assistant use your account directly: list voices and models, turn text into speech and keep pronunciation dictionaries. You sign in with your KugelAudio account; there is no API key to copy.

When one of your organizations has speech to text, the assistant can also transcribe recordings with Kugel Ear 1.

Text
https://www.kugelaudio.com/mcp

Connect

claude mcp add --transport http kugelaudio https://www.kugelaudio.com/mcp

In claude.ai, add the URL above as a custom connector. Any client that supports remote MCP servers with OAuth works the same way. When the client connects, your browser opens a KugelAudio sign-in page: press Allow, and the assistant works in every project you manage.

What the assistant can do

ToolWhat it does
account_whoamiThe account the assistant acts as, every organization and project it works in, and the billing project
account_set_billing_projectMoves speech billing to another project you manage, when you tell the assistant to
speech_list_modelsText-to-speech models the billing project's organization can use
speech_list_voicesVoices available in the billing project, optionally by language
speech_synthesizeSpeech from text in a voice from speech_list_voices, billed to the billing project; returns a one-time link to a WAV file. dictionary_ids picks the dictionaries it applies, as in the API

When one of your organizations has access to Kugel Ear 1, account_whoami says so and two more tools appear; they transcribe in the billing project, whose organization needs the access too, and speech_list_models then also lists kugel-ear-1:

ToolWhat it does
speech_transcribeTranscript of a short clip, up to 10 MiB, that the assistant sends inline
speech_create_transcription_upload_linkA one-time link the assistant can curl a recording to (curl -sS -F file=@call.wav <link>) and get its transcript back as JSON

Both accept boosted phrases, the custom vocabulary of speech to text.

Links from speech_synthesize are valid for 15 minutes and work once. A speech link synthesizes when it is downloaded, so an assistant with a shell saves the file with curl -sS -o speech.wav <link>; in a chat app, open the link to download it. Write the text the way it should be spoken, as in Generate speech.

Links from speech_create_transcription_upload_link are valid for 15 minutes and work once, too.

Documentation

Every connection can search and read these docs and the Curb docs, so a coding agent can look things up while it builds:

ToolWhat it does
search_docsPages and passages that match a query, each with its site and URL. Searches both sites, or one with site set to kugelaudio or curb
read_docs_pageOne page by its URL or path: its site, URL, title and Markdown. A path means these docs unless site is curb

They include the pages of models your organizations have early access to.

Without an account, the same two tools are at https://docs.kugelaudio.com/mcp, with no sign-in and public pages only:

Bash
claude mcp add --transport http kugelaudio-docs https://docs.kugelaudio.com/mcp

Pronunciation dictionaries

Every connection can keep the billing project's pronunciation dictionaries, as the dashboard's Dictionaries page does:

ToolWhat it does
dictionary_listThe billing project's dictionaries, with their language and whether they are active
dictionary_createA new dictionary
dictionary_updateRenames a dictionary, changes its description or language, or switches it on or off
dictionary_deleteDeletes a dictionary and its entries
dictionary_list_entriesA dictionary's entries, searched by word, or all of them to export
dictionary_add_entriesAdds entries, replacing the entry of a word that is already there, or replaces the whole list
dictionary_update_entryChanges one entry
dictionary_delete_entriesDeletes entries

Changes apply to the next request. A client connected before these tools existed is asked to reconnect when it first uses one, because its key cannot change dictionaries.

Billing and access

A connection works in every project you manage in your organizations: every project of an organization you own or administer, and each project you are an admin of. It follows your access on every call, so a project you start managing is included and one you stop managing drops out. An organization that requires two-factor sign-in is included only when you allowed the connection with two-factor sign-in; account_whoami names any organization left out and what to do.

Speech is billed to one of those projects, the billing project. Speech, voices, models and pronunciation dictionaries all act there, and every call is an ordinary API request in it, billed to its organization like any other usage (see pricing). With only one project, that project is the billing project. Otherwise the assistant asks you which project to bill before it first makes speech, and account_set_billing_project moves billing when you ask; the assistant never chooses for you.

The first time the assistant makes speech in a project, an API key named MCP: <assistant name> is created there, on that project's API keys page. It can generate speech, read voices and change the project's pronunciation dictionaries. No tool ever shows the key.

The keys can also transcribe audio, in an organization that has speech to text.

To disconnect an assistant, delete any one of its keys, or disconnect it in the client: either ends the whole connection, in every project. The assistant also stops working when you no longer manage any project.

Next steps

  • Voices to pick a voice for the assistant to use
  • Agent skill if you are building your own integration with a coding agent