MCP server

Connect Cursor, Claude Desktop, ChatGPT, or other MCP clients to the Speechdash public API.

Speechdash ships an official Model Context Protocol (MCP) server that exposes the library and speech synthesis as tools. Use it when you want an AI assistant to list documents (including finished transcripts), create text imports, translate or export, check your wallet, or synthesize speech without writing HTTP calls yourself. The library includes finished transcripts (source=transcription). There is no transcripts tool: read those jobs with the document tools. MCP does not start audio transcription jobs.

You can also call the public REST API directly with curl or your own client. MCP and HTTP share the same endpoints, credits, and rate limits where both surfaces exist.

Inbound (this page): Speechdash's official MCP at mcp.speechdash.com lets external assistants call your Speechdash library.

Outbound (app settings): Settings → Connectors attaches third-party MCP servers to the in-app chat UI when that feature is enabled. That is not Speechdash's public MCP server.

Never set SPEECHDASH_API_KEY on the public mcp.speechdash.com deployment. Each client must authenticate with OAuth or a personal sh_live_… key in the Authorization header.

Before you start

Create an API key under Settings → API in the app for Cursor, Claude Desktop, and other clients that accept bearer keys. MCP uses the same sh_live_… keys as HTTP. See API overview.

ChatGPT (OAuth)

ChatGPT requires OAuth 2.1 + PKCE, not a pasted API key. Do not enter an API key in the ChatGPT connector UI.

MCP URL: https://mcp.speechdash.com/mcp
OAuth resource: https://mcp.speechdash.com
Protected resource metadata: https://mcp.speechdash.com/.well-known/oauth-protected-resource
Authorization server (discovery): https://speechdash.com/.well-known/oauth-authorization-server

Sign in with your Speechdash account when ChatGPT prompts you. Access tokens are minted by the app and accepted by the MCP server and public API.

Setup walkthrough: Agents hub.

Tools

Each MCP tool maps to one REST endpoint:

MCP tool	REST
`speechdash_get_account`	`GET /v1/me`
`speechdash_list_documents`	`GET /v1/documents`
`speechdash_get_document`	`GET /v1/documents/{document_id}`
`speechdash_create_document`	`POST /v1/documents`
`speechdash_update_document`	`PUT /v1/documents/{document_id}`
`speechdash_translate_document`	`POST /v1/documents/{document_id}/translate`
`speechdash_export_document`	`GET /v1/documents/{document_id}/export`
`speechdash_delete_document`	`DELETE /v1/documents/{document_id}`
`speechdash_list_voices`	`GET /v1/voices`
`speechdash_get_voice`	`GET /v1/voices/{voice_id}`
`speechdash_synthesize_speech`	`POST /v1/audio/speech`

The server also exposes resource speechdash://openapi (live OpenAPI JSON).

MCP vs REST capabilities

Feature	MCP	REST
Account snapshot	Yes (`speechdash_get_account`)	`GET /v1/me`
Document CRUD (library, including finished transcripts)	Yes	Yes
Start audio transcription jobs	No (app only)	No (app only)
Non-streaming speech (5k chars)	Yes	`POST /v1/audio/speech`
Streaming speech (20k chars)	No, use REST	`POST /v1/audio/stream`
Streaming with timestamps	No, use REST	`POST /v1/audio/stream/with-timestamps`
Translate	Yes (`speechdash_translate_document`)	`POST /v1/documents/{document_id}/translate`
File export	Yes (`speechdash_export_document`)	`GET /v1/documents/{document_id}/export`
Archive / restore	Yes (`is_archived` on `speechdash_update_document`)	`PUT /v1/documents/{document_id}`
Health check	No	`GET /health`
OpenAPI JSON	Resource `speechdash://openapi`	`GET /v1/openapi.json`

Billing

MCP speech synthesis debits the same credit wallet as REST and cloud playback in the app. Creating documents via MCP does not spend credits, but it counts toward the same daily document cap as the app and REST API. Call speechdash_get_account before synthesis to check your balance.

speechdash_translate_document bills the same credit wallet as MP3 export (0.5 display credits / 1 wallet cent per started 30 seconds of estimated speech). Transcript clocks (**0:34**, **SPEAKER_00** · 0:34, (0:34)) are copied unchanged as playback references. speechdash_export_document can return large file bytes: prefer txt, csv, or srt in MCP. pdf and docx are binary. speechdash_update_document accepts visibility for the unlisted share page and is_archived to archive or restore.

Cursor or Claude Desktop

Use Streamable HTTP against the hosted MCP server with your personal API key. Example .cursor/mcp.json:

{
  "mcpServers": {
    "speechdash": {
      "url": "https://mcp.speechdash.com/mcp",
      "headers": {
        "Authorization": "Bearer sh_live_your_key_here"
      }
    }
  }
}

Create the key under Settings → API in the app.

Local development (repository clone)

If you work from a clone of the Speechdash repository:

export SPEECHDASH_API_KEY="sh_live_..."
bun run dev:mcp

Optional: point at a local API with SPEECHDASH_API_BASE=http://localhost:3003/v1 while bun dev:api is running.

Streamable HTTP (local)

For HTTP MCP clients during local development:

export SPEECHDASH_API_KEY="sh_live_..."
bun run dev:mcp:http

Default endpoint: http://127.0.0.1:3010/mcp

Variable	Purpose
`SPEECHDASH_API_KEY`	Required for stdio; optional on a local HTTP server if clients send `Authorization: Bearer`
`SPEECHDASH_API_BASE`	API base URL (default `https://api.speechdash.com/v1`)
`MCP_TRANSPORT`	`stdio` (default) or `http`
`MCP_HOST`	HTTP bind host (default `127.0.0.1`)
`MCP_PORT`	HTTP port (default `3010`)

Before you start

ChatGPT (OAuth)

Tools

MCP vs REST capabilities

Billing

Cursor or Claude Desktop

Local development (repository clone)

Streamable HTTP (local)

Agent-readable docs

Agents hub (website)

agents.md

OpenAPI JSON

Examples on GitHub

API overview

API keys and MCP (user guide)

Credit wallet

Connectors (app settings)

On this page