Connect AI Assistants with MCP
Acoust supports the Model Context Protocol (MCP) so an AI assistant can find voices, check your character credits, and generate short speech clips on your behalf. You connect with your Acoust account through OAuth; you do not need an API key or a local Acoust server.
Before You Connect
You need an Acoust account, free or paid, plus an assistant that supports remote MCP servers over Streamable HTTP and OAuth sign-in with dynamic client registration. Availability and setup controls depend on your assistant; support for MCP alone does not guarantee compatibility with Acoust.
Generation uses your existing personal or team character credits. Free personal accounts can connect and generate within their remaining allowance; connecting does not add credits. Team members still need active access to their team's allowance. REST API access has a separate paid-plan requirement. Review Billing and Usage if you need to check your allowance first.
Connect Your Assistant
-
Open your assistant's MCP server or connector settings and add a remote server.
-
Give the connection a recognizable name, such as Acoust, and enter this server URL:
https://api.acoust.io/mcp -
Choose OAuth authentication if your assistant asks for an authentication method, then start the connection.
-
Sign in to Acoust when prompted. On Connect an application, review the application name, redirect destination, account, requested permissions, and usage pool. Continue only if you recognize the application and its destination.
-
Select Approve to authorize access, or Deny to cancel. Return to your assistant to finish connecting.
The permissions let an assistant list available voices, read usage, and generate speech, depending on what it requests. Approving speech generation lets the assistant consume your shared credits.
If the authorization page expires, start a new connection from your assistant. Do not reuse an old authorization link.
Try Your First Requests
Start by discovering voices and checking usage. Neither action generates audio or consumes your speech character credits.
Show me the Acoust voices available in US English and their supported controls. Do not generate audio yet.
Check my remaining Acoust character credits and tell me whether they are personal or shared with my team.
Once you have picked a voice, request one clip:
Using the voice I selected and US English, generate one Acoust speech clip saying: "Welcome to our weekly update." Give me the MP3 download link. Do not automatically retry if generation fails or times out.
Your assistant uses these tools behind the scenes:
Swipe or scroll sideways to read all columns. With the table focused, use the left and right arrow keys.
| Tool | What it does |
|---|---|
list_voices | Lists available voices, their IDs, languages, supported controls, and input limits. Includes your own approved cloned voices. |
get_usage | Reads your shared character usage, remaining credits, and reset time when available. It does not reserve credits. |
generate_speech | Generates one MP3 clip and returns a download link, expiry time, and character charge. |
The assistant should use voice IDs returned by list_voices, rather than guessing names or IDs. For a voice with multiple languages, it must select one of the returned language codes. Speed and delivery instructions are available only when the selected voice lists those controls.
Limits, Usage, and Downloads
- Each generation accepts plain text, up to 4,000 characters after trimming surrounding whitespace. Counting uses UTF-16 units, so some symbols, including emoji, count as more than one. Individual voices may have lower character or byte limits; ask your assistant to check the voice's limits before submitting long text. SSML and HTML are not interpreted as speech markup.
- Generating speech with a standard voice uses one credit per input character. Generating speech with a cloned voice uses twice the input character count. For example, a 100-character generation with a clone uses 200 credits.
- Team members use their team's shared credits. If your usage pool changes, reconnect and review the new pool before generating again.
- Download and save the MP3 within 24 hours. Downloading an existing result does not generate speech again. Expired links cannot be renewed through MCP, and MCP does not provide generation history or result recovery.
- MCP supports short speech generation. Create a clone by recording in the Voice Cloning workspace; MCP can use your existing approved clones but cannot create clones or edit Acoust projects.
A timeout, disconnect, or missing result does not prove generation failed. Audio may have been generated and credits consumed. Acoust does not automatically retry generation, and your assistant should not repeat a call after an error or reconnect. Request another generation only when you choose to proceed, understanding that it may consume credits again.
Manage or Disconnect an Assistant
- Open API & Integrations in Acoust.
- Select Connected Apps (MCP) to see each application's permissions, usage pool, and connection date.
- Select Revoke beside the application, then Confirm revoke.
Revoking blocks future requests from that connection. Requests already submitted may still finish, and previously issued audio links remain usable until they expire. Other connections and REST API keys are unaffected. To use the assistant again, start a new OAuth connection from that assistant.
Troubleshooting
Swipe or scroll sideways to read all columns. With the table focused, use the left and right arrow keys.
| Problem | What to do |
|---|---|
| The assistant asks for an API key or cannot complete sign-in | Check that it supports remote Streamable HTTP, OAuth, and dynamic client registration. Use https://api.acoust.io/mcp; REST API keys do not authenticate MCP connections. |
| Authorization expired, was denied, or the connection was revoked | Start a new OAuth connection from the assistant and approve access with the intended Acoust account. Reconnecting should not repeat an earlier generation. |
Your team access is no longer available (plan_required) | Ask your team administrator to confirm your membership and the team's active allowance. Free personal accounts do not need a paid plan for MCP. |
Your billing or usage pool changed (consent_required) | Reconnect and review the personal or team usage pool shown on the consent page. |
| A voice, language, or setting is rejected | Refresh the voice list and use a returned voice ID, supported language, and supported controls. Shorten text if it exceeds the voice's limits. |
Your character credits are exhausted (quota_exceeded) | Check usage and the reset time, or review your plan. Team usage includes other members' generation. |
Too many requests (rate_limit_exceeded) | Pause new generation requests and let active work finish. If a previous result is uncertain, do not automatically resubmit it. |
| Speech generation is missing | Confirm the connection includes speech generation permission. Tool availability can depend on the authorization and current service configuration. If the problem persists, contact support with the assistant name, time, and error details. |
| Generation failed, timed out, or disconnected | Do not automatically retry. If a response arrived, check generation_outcome and quota_status; only refunded confirms a refund, and unknown means the charge or outcome could not be confirmed. |
| The audio link has expired | Check whether you saved the MP3. A replacement requires a new generation request and may consume credits again. |
For help, email hi@acoust.io with your assistant's name and version, the time of the issue, any error code, and the request_id if available. Do not include passwords, access tokens, authorization codes, or private audio links.
Building an application with API keys instead? See the REST API guide.