Conversions across 221 color spaces, tested against cited sources: CSS strings, gamut, LUTs
Cartesia is installed from its publisher's own source and answers where it runs, so this marketplace is not in the path of a single call. There is no address here to send one to, and a panel that pretended otherwise would be showing you an answer we made up. Install it and call it from your own client — the Installation tab has the entry for each one.
What it does
Cartesia voice AI: text-to-speech, speech-to-text, voice cloning, and voice management.
Quickstart
# 1 — add it to Claude Code
claude mcp add --scope user --env 'CARTESIA_API_KEY=${CARTESIA_API_KEY}' --transport stdio cartesia-ai-cartesia-mcp '--' uvx cartesia-mcp
# 2 — ask your agent for something these tools do
# text_to_speech, localize_voice, delete_voice and 17 more
It needs CARTESIA_API_KEY: this reads it from your environment, so export it first. Remote: OAuth sign-in · Package: API key.
Cartesia costs nothing on mcprush: there is no plan to choose here, no cap to set and nothing mcprush can bill you for.
Collected from a public index. Nobody has claimed this account, so nothing here was written by its author — claim it if it is yours.
Where are you running it?
Every route below installs the same thing and ends at the same approval screen. This one can run either way — at an address its publisher runs, or as a process under your own user, with your files and your network — and this marketplace is in the path of neither.
This is a public server: you run it yourself and this marketplace is not in the path. You need uv (docs.astral.sh/uv): uvx comes with it. Claude Code registers it in one command; --scope user makes it available in every project. Claude Code reads CARTESIA_API_KEY (Cartesia API key from https://play.cartesia.ai/keys) from your environment each time it starts the server, so the value stays out of its config: export it in your shell profile (macOS, Linux), or run setx CARTESIA_API_KEY … and open a new terminal (Windows). The '--' is in quotes so that PowerShell passes it on.
claude mcp add --scope user --env 'CARTESIA_API_KEY=${CARTESIA_API_KEY}' --transport stdio cartesia-ai-cartesia-mcp '--' uvx cartesia-mcpReconnect, or start a new session, and the tools appear in the model’s tool list.
Its publisher also runs it at https://mcp.cartesia.ai/mcp: nothing to start on your machine, and this marketplace is not in the path.
claude mcp add --scope user --transport http cartesia-ai-cartesia-mcp https://mcp.cartesia.ai/mcpIt reaches a system of yours, so it needs your own credential rather than the publisher’s: CARTESIA_API_KEY — Cartesia API key from https://play.cartesia.ai/keys. You set it in your own client’s config, on your machine — this marketplace never holds it.
One config entry your client uses to start the process locally. A local server runs with your file system and your network, which is why it is priced without metering.
20 tools, with what each one reads, writes and reaches shown before you agree — the same list on every route above. Read the tool surface.
Tool surface
What the model actually sees. Descriptions are diffed on every release — see version history.
Read from its source code. A server that builds its tool list when it starts can list more tools than this.
Generate speech audio from text. By default (save=true) the audio is persisted in Cartesia cloud storage and the response includes file_id and download_url (24-hour public link). Pass that file_id to speech_to_text or clone_voice — including from a hosted client…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| transcript | string | yes | Text to synthesize. |
| voice_id | string | yes | Cartesia voice ID (e.g. from list_voices). |
| output_format | yes | Audio output format: container (wav, mp3, or raw), encoding, and sample_rate. | |
| model_id | string | no | The ID of the model to use for the generation. Call list_models with capability="tts" for valid ids. |
| language | no | Language of the transcript as an ISO 639-1 code (e.g. en, es). | |
| duration | number | no | The maximum duration of the audio in seconds. You do not usually need to specify this. If the duration is not appropriate for the length of the transcript, the output audio may be truncated. |
| speed | number | no | Speech speed multiplier (0.6–1.5). Applies to Sonic 3+ via generation_config. |
| volume | number | no | Volume multiplier (0.5–2.0). Applies to Sonic 3+ via generation_config. |
| emotion | string | no | Emotional guidance for the generation (e.g. neutral, excited, sad). Applies to Sonic 3+. |
| pronunciation_dict_id | string | no | Pronunciation dictionary ID to apply for this generation only. |
| output_path | string | no | Local file path to write the audio (for example src/assets/audio/click.wav). Relative paths resolve against the MCP server working directory. Only supported with the local server (stdio / uvx); reje |
| save | boolean | no | When true (default), persist the generation to cloud storage and return file_id plus download_url. Pass download_file with that file_id to refresh the link or write another local copy. |
Create a new voice from an existing voice localized to a new language and dialect.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| voice_id | string | yes | The ID of the voice to localize. |
| name | string | yes | The name of the new localized voice. |
| description | string | yes | The description of the new localized voice. |
| accent | yes | Catalog accent id from GET /accents (for example mexican or parisian). Display names are rejected on this API version. | |
| language | yes | Target language to localize the voice to. Distinct from TTS SupportedLanguage. | |
| original_speaker_gender | yes | Gender of the speaker in the source voice. | |
| dialect | no | Dialect allowlist for English, Spanish, Portuguese, and French. |
Permanently delete a voice you own.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| voice_id | string | yes | The ID of the voice to delete. |
Get a voice by ID.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| voice_id | string | yes | The ID of the voice to get. |
Update a voice's name and description.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| voice_id | string | yes | The ID of the voice to update. |
| name | string | yes | The name of the voice. |
| description | string | yes | The description of the voice. |
List the official accent catalog from GET /accents. Each entry includes the canonical accent id plus display metadata (name, language, locale, is_locale_default, is_localizable). Pass id to clone_voice, add_voice_accents, delete_voice_accent,…
Takes no parameters.
List callable Sonic (TTS) and Ink (STT) model ids from GET /models. Each row has id and capability (tts or stt). TTS rows include supported_languages, the sorted synthesis language codes that model accepts. STT rows include modes (stream and/or…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| capability | no | Filter to TTS or STT. Omit to return both. |
List LLMs available for Managed Agents from GET /v1/agents/models. Use these id values for an agent's model.id (and explicit fallback model ids). Do not invent model ids. Each row includes provider, display name, latency, and token pricing. Returns the full…
Takes no parameters.
Add accents to an instant voice clone you own so a single voice_id can speak multiple accents natively. A voice can support up to 10 accents in total, including native.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| voice_id | string | yes | The ID of the voice. |
| accents | array | yes | Catalog accent ids from GET /accents (for example british or parisian). Display names are rejected on this API version. |
Remove an accent an instant voice clone supports.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| voice_id | string | yes | The ID of the voice. |
| accent | yes | Catalog accent id from GET /accents (for example british or parisian). Display names are rejected on this API version. |
Clone a voice from an audio clip. This endpoint has two modes, stability and similarity. Similarity mode clones are more similar to the source clip, but may reproduce background noise. For these, use an audio clip about 5 seconds long. Stability mode clones are…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| name | string | yes | The name of the voice. |
| language | yes | The language of the voice. | |
| mode | string | yes | Tradeoff between similarity and stability. Similarity clones sound more like the source clip, but may reproduce background noise. Stability clones always sound like a studio recording, but may not sou |
| file_path | string | no | Absolute path on this MCP server. Must exist; hosted MCP only accepts files this server wrote (from download_file or save=false TTS). |
| file_id | string | no | Cartesia cloud file id. The server downloads it. |
| description | string | no | A description for the voice. |
| accent | no | Catalog accent id from GET /accents (for example southern-us or parisian). Display names are rejected on this API version. Omit to leave unset. |
List voices available to you (your own and Cartesia library voices), with search, filters, and cursor pagination.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| limit | integer | no | The number of Voices to return per page, ranging between 1 and 100. |
| starting_after | string | no | A cursor to use in pagination. starting_after is a Voice ID that defines your place in the list. For example, if you make a /voices request and receive 100 objects, ending with voice_abc123, your |
| ending_before | string | no | A cursor to use in pagination. ending_before is a Voice ID that defines your place in the list. For example, if you make a /voices request and receive 100 objects, starting with voice_abc123, your |
| is_owner | boolean | no | Whether to only return voices owned by the current user. |
| is_starred | boolean | no | Whether to only return starred voices. |
| gender | no | The gender presentation of the voices to return. | |
| language | string | no | Filter voices by language or locale, such as en, it, or en_GB. A locale returns accents for that region; a language alone returns all accents for that language. Both - and _ separators are a |
| q | string | no | Search voices by name, description, or voice ID. |
| expand | array | no | Additional fields to include in the response, such as preview_file_url. |
Transcribe a pre-recorded audio file to text. Default (mode="batch"): upload the file via batch STT (POST /stt). Best for typical MCP file-on-disk tasks (mp3, flac, wav, and other containers). Default model: ink-whisper. **Streaming…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| file_path | string | no | Absolute path on this MCP server. Must exist; hosted MCP only accepts files this server wrote (from download_file or save=false TTS). |
| mode | no | batch (default) or stream. | |
| model | string | no | STT model ID. Defaults to ink-whisper for batch and ink-2 for stream. Call list_models with capability="stt" for valid ids. |
| language | string | no | ISO-639-1 language code (defaults to en for stream; batch uses API default). |
| encoding | no | Required for raw PCM without a container header (both modes). | |
| sample_rate | integer | no | Sample rate in Hz when encoding is set. |
| timestamp_granularities | array | no | Pass ["word"] for word-level timestamps when supported. |
| file_id | string | no | Cartesia cloud file id. The server downloads it. Use this from hosted clients. |
Fetch a Cartesia cloud-stored file by ID. Returns a 24-hour download_url for hosted and remote clients. Also writes a copy to OUTPUT_DIRECTORY on the MCP server host (for local uvx or server-side speech_to_text in the same session). On hosted MCP those local copies…
| Parameter | Type | Required | Notes |
|---|---|---|---|
| file_id | string | yes | Cloud file ID (e.g. from text_to_speech). |
| format | no | Pass playback to wrap raw PCM as WAV in the download link and local file. |
Returns credit usage over time (GET /usage/credits). Requires CARTESIA_ADMIN_API_KEY (admin API keys cannot be used as CARTESIA_API_KEY). Optional start_ts and end_ts are RFC 3339 datetimes; interval buckets results by day, week, or month.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| start_ts | string | no | Start of the usage window (ISO 8601 timestamp). |
| end_ts | string | no | End of the usage window (ISO 8601 timestamp). |
| interval | no | Bucket size for the returned usage time series. | |
| api_key_id | string | no | Limit usage to a specific API key ID. |
| group_by | no | Break down usage by capability, model, voice, or API key. When interval is set it must be day. Grouped responses use { group_by, data: [{ id, label?, buckets }] }. |
List pronunciation dictionaries for the authenticated user.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| limit | integer | no | Number of dictionaries per page (1–100). |
| starting_after | string | no | Cursor: dictionary ID to start after. |
| ending_before | string | no | Cursor: dictionary ID to end before. |
Create a pronunciation dictionary.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| name | string | yes | Name of the pronunciation dictionary. |
| items | array | no | Mappings of text to pronunciation (IPA or sounds-like). Optional case_sensitive (default false): when false, match every capitalization of text on Sonic 3.6. |
Get a pronunciation dictionary and its items by ID.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| dict_id | string | yes | Pronunciation dictionary ID. |
Update a pronunciation dictionary.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| dict_id | string | yes | The ID of the pronunciation dictionary to update. |
| name | string | no | New name for the pronunciation dictionary. |
| items | array | no | Optional case_sensitive (default false) on each item. |
Permanently delete a pronunciation dictionary.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| dict_id | string | yes | Pronunciation dictionary ID to delete. |
- Every tool, no call limit
- No card, and mcprush charges nothing for it
- Source is public, though no licence is declared for it
- Runs on your machine — nothing of it reaches our gateway
- Nothing to cap, because mcprush bills nothing
What counts against your monthly calls
| Tool | Unit | Calls used | Out of the allowance |
|---|
Nothing here is billable: mcprush charges nothing to install Cartesia or to call it, at any volume.
Two independent axes, because powerful and malicious are different questions. The grade is threat only. The capability level is blast radius, and it is never a penalty on the grade — it is priced as one subtract-only term in the score, where you can see it.
| Term | Level | What it prices | Points |
|---|---|---|---|
| capability-exposure | minimal | capability blast radius (minimal) — client exposure if the model is manipulated | −0 |
| verification-discount | none | publisher verification (unlinked) — no provenance/repo link, but the shipped source was fully read | −2 |
| coverage-honesty | source | inspection depth (source) — how much of the target the scan could see | −0 |
What the scan could actually read
A grade is only as meaningful as its coverage, so the scanner publishes its own depth before it publishes its result.
Tools were statically extracted from the published source (20 recovered), not enumerated from a running server. Tool-poisoning, Unicode-smuggling, capability and toxic-flow analysis ran on this inferred surface, but a mis-parsed registration could be missed or mis-attributed, so tool-derived findings are capped below “confirmed”. To grade the real runtime surface, scan the running server: --command "npx -y <package>".
The source has registrations with unknown names: the tools registered at cartesia_mcp/mcpserver.py:63 get their names only at run time (a loop over a list the code builds, a name it computes), so they are not in this surface and were not analysed. To see them, scan the running server: --command "npx -y <package>".
Capability — what it could do if the model were manipulated
Tags derived from each tool’s schema and the implementation, not from what the tool calls itself. minimal is the level these add up to.
| Tool | Capability tags | Why the tag was assigned |
|---|---|---|
| text_to_speech | no tags | |
| localize_voice | no tags | |
| delete_voice | no tags | |
| get_voice | no tags | |
| update_voice | no tags | |
| list_accents | no tags | |
| list_models | no tags | |
| list_agent_models | no tags | |
| add_voice_accents | no tags | |
| delete_voice_accent | no tags | |
| clone_voice | no tags | |
| list_voices | no tags | |
| speech_to_text | no tags | |
| download_file | untrusted-input | untrusted-input: keyword "download" |
| get_credit_usage | no tags | |
| list_pronunciation_dicts | no tags | |
| create_pronunciation_dict | no tags | |
| get_pronunciation_dict | no tags | |
| update_pronunciation_dict | no tags | |
| delete_pronunciation_dict | no tags |
Toxic-flow graph
The lethal trifecta, checked as a graph rather than as a checklist: untrusted input, a sensitive source and an external sink have to meet before there is a path worth worrying about.
The public result for this release does not print the flow graph, so there is nothing to show here. That is not the same as "no paths were found": what the scan did read is above, under coverage.
Supply chain and provenance
This is the first scan of this surface here, so there is nothing yet to compare it against.
Every result on this tab comes from one deterministic pass over the published package — offline, rule by rule, and auditable line by line above. Same methodology version, same bytes, same score.
Release history
Always latest — the install command names no version, so it installs the newest release when it runs; the current release on file is 0.34.1. Pin a release below and the command asks for exactly that one.
No release note was published with this version.
No release note was published with this version.
- The version this listing was on when mcprush began following its releases. The registry was not asked when it shipped, so this row carries no date.
One review per account, from a signed-in account that does not publish this listing. Nothing else is asked — you can say what you think before you install it. Publishers can reply once per review, and a review can be edited or deleted by whoever wrote it.
Any signed-in account that does not publish Cartesia can review it, once — before installing it or after. A review from an account that has installed it is marked as one.
Nobody has reviewed this listing. The rating on the card is the mean of the reviews written here and nothing else, so there is no rating until somebody writes the first — which takes a signed-in account that does not publish it, and nothing else.