Speech and audio APIs and MCP servers for AI agents

Text-to-speech, speech-to-text, voice cloning, dubbing and audio intelligence.

13 speech and audio APIs an AI agent can call: auth method and header, official MCP server (8), free tier (6), docs links. Machine-readable JSON included. Agents: fetch catalog/audio-speech.json instead of parsing this page. Last checked 2026-09-24.

Ratings and usage (★) fetched: 2026-09-24 16:54 UTC – 2026-09-24 17:02 UTC.

Service	What an agent can do	Auth	MCP server	Free tier	Details (with sources)	Notes	Proof (ratings, usage)
Amazon Polly / Amazon Transcribe (AWS)	Convert text to lifelike speech (Polly) and transcribe audio with speaker diarization (Transcribe) via AWS API.	cloud IAM `AWS SigV4 signing with IAM access keys`	official	Free tier: Polly 5M chars/mo standard (neural 1M, generative 100K chars/mo for first 12 months), Transcribe 60 min/mo (12 months); new AWS accounts since 2025-07-15 get up to $200 Free Tier credits	Base URL: https://polly.<region>.amazonaws.com Free plan: Polly: 5M standard chars/month (neural 1M for first 12 months); Transcribe: 60 min/month for 12 months, card required ↗ Data policy: Transcribe may store and use voice inputs to improve AWS AI services; opt out via AWS Organizations AI services opt-out policy ↗ Async jobs: Transcribe: StartTranscriptionJob, then poll GetTranscriptionJob ↗ Example call: synthesize speech from text (Polly): POST https://polly.{region}.amazonaws.com/v1/speech ↗ Unit price: Polly: standard $4.00, neural $16.00, generative $30.00, long-form $100.00 per 1M characters (beyond free tier) ↗ MCP tools: 3: call_aws, suggest_aws_commands, get_execution_plan ↗	Needs an AWS account and IAM credentials. Voice and feature availability differs by region. Transcribe docs: https://docs.aws.amazon.com/transcribe/latest/dg/what-is.html. The generic AWS API MCP Server (awslabs) runs AWS CLI commands, incl. polly and transcribe.	—
AssemblyAI API	Transcribe recorded or streaming audio with speaker diarization, sentiment analysis and LLM-based audio insights via REST.	API key `Authorization: <key> (no Bearer prefix)`	—	Trial: $50 in free credits one-time on signup, no card required, credits do not expire (LLM Gateway not included); then pay-as-you-go	Base URL: https://api.assemblyai.com Free plan: $50 credits (do not expire), one-time, no card ↗ Rate limits: Pre-recorded: 5 parallel transcriptions on free accounts, 200+ on paid; excess jobs are queued FIFO ↗ Data policy: Certain API files may be used for model training after redaction unless opted out; zero data retention for Streaming when opted out ↗ Async jobs: POST /v2/transcript returns id; poll GET /v2/transcript/{id} or set webhook_url ↗ Example call: transcribe an audio file: POST https://api.assemblyai.com/v2/transcript ↗ Unit price: Pre-recorded speech-to-text: Universal-3.5 Pro $0.21 per audio hour, Universal-2 $0.15 per hour ↗ MCP: docs-only: the server searches documentation, it cannot call the API ↗	Vendor MCP at https://assemblyai.com/docs/mcp only searches the docs. To run transcriptions, call the REST API or an SDK.	—
Azure AI Speech	Convert text to speech, transcribe speech to text and translate speech via REST or Speech SDK.	cloud IAM `Ocp-Apim-Subscription-Key: <key>, or Authorization: Bearer <Entra token>`	official	Free tier (F0): 5 audio hrs/mo STT, 0.5M chars/mo neural TTS; batch transcription not available on F0	Free plan: 5 audio hours STT + 0.5M neural TTS characters, per month ↗ Rate limits: F0 quotas are not adjustable; F0 TTS 20 transactions per 60 seconds ↗ Example call: convert text to speech: POST https://{region}.tts.speech.microsoft.com/cognitiveservices/v1 ↗ Unit price: Pay-as-you-go (East US, S1): neural text to speech $15 per 1M characters; real-time speech to text $1 per audio hour ↗ MCP tools: 2: speech stt recognize, speech tts synthesize ↗	Now branded 'Azure Speech in Foundry Tools'. Needs an Azure Speech resource, and endpoints are per region. Speech tools (speech stt recognize, speech tts synthesize) ship in the Azure MCP Server.	GitHub 1★ microsoft/azure-speech-mcp-server (2026-09-24 16:54 UTC)
Cartesia API	Generate low-latency real-time text-to-speech, transcribe speech and clone voices via REST/WebSocket API.	API key `Authorization: Bearer <key> plus Cartesia-Version header`	official	Free tier: 20,000 credits/mo; no commercial-use license and no voice cloning	Base URL: https://api.cartesia.ai Free plan: 20K credits (~1 credit per TTS character), per month ↗ Rate limits: Free plan: 2 concurrent TTS requests, 8 concurrent STT requests; limits scale with subscription plan ↗ Data policy: Zero Data Retention available to Enterprise customers for TTS and STT ↗ Example call: text-to-speech (audio file): POST https://api.cartesia.ai/tts/bytes ↗ Unit price: TTS: about 1 credit per character; Pro plan $5/mo for 100K credits, Startup $49/mo for 1.25M credits ↗ MCP tools: 16: text_to_speech, speech_to_text, list_voices, get_voice, clone_voice, update_voice … ↗	The hosted MCP endpoint uses OAuth. The local MCP server is at https://github.com/cartesia-ai/cartesia-mcp.	—
Deepgram API	Transcribe batch or streaming audio with diarization, synthesize speech, and run real-time voice agents via REST/WebSocket.	API key `Authorization: Token <key>`	official	Trial: $200 in free credit one-time on signup, no credit card required; then pay-as-you-go	Base URL: https://api.deepgram.com Free plan: $200 credit, one-time, no card ↗ Rate limits: Nova-3/Nova-2: up to 50 concurrent pre-recorded and 150 streaming requests; Voice Agent API up to 45 concurrent connections ↗ Data policy: Model Improvement Program may use data for training; opt out per request with mip_opt_out=true ↗ Example call: transcribe a pre-recorded audio file from a URL: POST https://api.deepgram.com/v1/listen ↗ Unit price: Nova-3 monolingual pre-recorded transcription: $0.0043 per audio minute (pay-as-you-go) ↗ MCP: docs-only: the server searches documentation, it cannot call the API ↗	Hosted MCP at https://developers.deepgram.com/_mcp/server only searches the docs; the API-calling server is local `dg mcp` (Deepgram CLI) or pip deepgram-mcp.	GitHub 2★ deepgram/mcp (2026-09-24 16:54 UTC)
ElevenLabs API	Generate, clone and stream text-to-speech voices and transcribe audio via REST/WebSocket API.	API key `xi-api-key: <key>`	official	Free: limited monthly credits; non-commercial use with attribution only, no voice cloning	Base URL: https://api.elevenlabs.io/v1 Free plan: free_tier, per month ↗ Rate limits: Free plan concurrency: 2 (Multilingual v2), 4 (Flash), 8 STT, 6 realtime STT; excess requests are queued ↗ Example call: convert text to speech: POST https://api.elevenlabs.io/v1/text-to-speech/{voice_id} ↗ Unit price: TTS API pay-as-you-go: $0.10 per 1,000 characters (Multilingual v2 / v3), $0.05 per 1,000 (Flash / Turbo) ↗	Official hosted MCP https://api.elevenlabs.io/v1/mcp uses OAuth; the local stdio repo elevenlabs/elevenlabs-mcp is archived since 2026-08-20.	GitHub 1,537★ elevenlabs/elevenlabs-mcp (2026-09-24 16:54 UTC) SourceForge 4.0★/4 reviews (2026-09-24 16:56 UTC)
Google Cloud Text-to-Speech / Speech-to-Text	Synthesize lifelike speech from text and transcribe audio to text via REST/gRPC using GCP credentials.	cloud IAM `API key (?key=) or OAuth2 service-account token: Authorization: Bearer <token>`	official	Free tier: TTS 4M chars/mo Standard and WaveNet, 1M chars/mo Neural2, Chirp 3 HD and Studio; STT 60 min/mo	Base URL: https://texttospeech.googleapis.com/v1 Free plan: TTS: 4M chars Standard/WaveNet, 1M Neural2/Chirp 3 HD/Studio; STT: 60 min, per month ↗ Data policy: STT offers a cheaper 'with data logging' price tier (opt-in); 'without data logging' tier costs more ↗ Example call: synthesize speech from text: POST https://texttospeech.googleapis.com/v1/text:synthesize ↗ Unit price: TTS after free tier: Standard and WaveNet $4, Neural2 $16, Chirp 3 HD $30, Studio $160 per 1M characters ↗ MCP tools: 2: chirp_tts, list_chirp_voices ↗	Requires a GCP project with billing enabled. Speech-to-Text docs are separate: https://cloud.google.com/speech-to-text/docs. Experimental vendor MCP mcp-chirp3-go (Genmedia) covers Chirp 3 HD TTS only.	—
Hume AI API	Generate expressive text-to-speech and run real-time empathic voice conversations (EVI) via REST/WebSocket.	API key `X-Hume-Api-Key: <key> (REST); api_key query param (EVI WebSocket)`	official	Free tier: 10,000 TTS chars/mo + 5 EVI min/mo; new accounts also get $20 in credits	Base URL: https://api.hume.ai/v0 Free plan: 10,000 TTS characters (~10 min); new accounts also get $20 credits, per month ↗ Rate limits: Free plan TTS: 15 requests per minute ↗ Data policy: By default anonymized EVI interaction data is used for training; users can opt out and enable 'Do not retain data' ↗ Example call: text-to-speech: POST https://api.hume.ai/v0/tts ↗ Unit price: TTS overage per 1,000 characters: $0.15 (Creator), $0.12 (Pro), $0.10 (Scale), $0.05 (Business); EVI $0.04-0.07/minute ↗ MCP: docs-only: the server searches documentation, it cannot call the API ↗ MCP tools: 5: tts, play_previous_audio, list_voices, save_voice, delete_voice ↗	The free plan allows 1 concurrent connection. Voice cloning is on paid plans. The API MCP is local `npx @humeai/mcp-server` (HUME_API_KEY); dev.hume.ai/_mcp/server only searches the docs.	—
Murf AI API	Generate AI voiceovers, clone voices and dub audio in 35+ languages via REST API.	API key `api-key: <key> request header (base URL https://api.murf.ai/v1)`	official	Trial: 100,000 characters one-time on the free API plan, no time limit	Base URL: https://api.murf.ai/v1 Free plan: 100,000 characters (no time limit), one-time ↗ Rate limits: Free tier streaming concurrency: Falcon 2 5 (US-East) or 2 (global/regional), Gen2 5; WebSocket connections 10x concurrency ↗ Example call: text-to-speech (generate a voiceover file): POST https://api.murf.ai/v1/speech/generate ↗ Unit price: Pay-as-you-go: $0.03 per 1,000 characters (minimum purchase $2); Falcon streaming model advertised at $0.01 per minute of speech ↗	Two model families: Falcon 2 (low latency, for streaming; $0.01/min) and Gen2 (studio quality).	SourceForge 5.0★/7 reviews (2026-09-24 16:56 UTC)
OpenAI Audio API (TTS / transcription)	Synthesize speech with TTS models and transcribe or translate audio with Whisper/GPT-4o transcribe via REST.	API key `Authorization: Bearer <key>`	—	paid only (pay-per-use)	Base URL: https://api.openai.com/v1 Rate limits: Usage tiers (Free, Tier 1-5) set rate limits and monthly usage caps; Free tier capped at $100/month ↗ Data policy: API data not used for training unless you opt in; abuse-monitoring logs retained by default (ZDR available to eligible customers) ↗	Whisper transcription costs about $0.006/min. TTS is billed per character or per token, depending on the model. Speech-to-text guide: https://developers.openai.com/api/docs/guides/speech-to-text.	GitHub 2,524★ openai/openai-openapi (2026-09-24 16:54 UTC)
Resemble AI API	Synthesize speech, clone voices, convert speech-to-speech and detect deepfake audio via REST API.	API key `API key from app.resemble.ai dashboard`	—	Paid only; Flex plan is $0/mo base with pay-as-you-go credits, no free usage allowance stated	Free plan: none, no card ↗ Rate limits: About 40 requests/second per API token; audio enhancement 10 requests/minute ↗ MCP: docs-only: the server searches documentation, it cannot call the API ↗	Enterprise plans are available for on-prem deployment and custom model training. Vendor MCP at https://docs.resemble.ai/_mcp/server only searches the docs.	SourceForge 1.7★/3 reviews (2026-09-24 16:56 UTC)
Rev AI API	Submit audio for asynchronous or streaming speech-to-text transcription, plus sentiment and topic extraction, via REST.	API key `Authorization: Bearer <access_token>`	—	trial credits (about 5 hours of ASR)	Base URL: https://api.rev.ai/speechtotext/v1 Free plan: Free credits equivalent to 5 hours of Reverb ASR, one-time ↗ Rate limits: Async STT: 10,000 transcription requests per 10 minutes per user; multipart uploads concurrency 5, 2 GB/file ↗ Data policy: Jobs and associated data deleted after max 30 days; configurable earlier auto-delete or delete_after_seconds ↗ Async jobs: POST /jobs returns job id; poll GET /jobs/{id} then GET /jobs/{id}/transcript, or use webhook notification ↗ Example call: submit an async transcription job: POST https://api.rev.ai/speechtotext/v1/jobs ↗ Unit price: Reverb ASR: $0.20 per hour English, $0.30 per hour other languages (per-second billing, 15 s minimum) ↗ MCP: docs-only: the server searches documentation, it cannot call the API ↗	Vendor MCP at https://docs.rev.ai/mcp only searches the docs. Human transcription is sold separately as a paid option.	—
Speechmatics API	Transcribe audio in batch or real time with speaker diarization across many languages via REST/WebSocket.	API key `Authorization: Bearer <key>`	community	Trial: one-time $100 credit grant, no payment card required	Base URL: https://eu1.asr.api.speechmatics.com/v2 Free plan: $100 credit grant, one-time, no card ↗ Rate limits: Batch: 10 new jobs/s, 50 job status requests/s; backstop 20,000 concurrent jobs; HTTP 429 when exceeded ↗ Async jobs: Batch: POST /v2/jobs, then poll GET /v2/jobs/{id} or use Notifications (callbacks) ↗ Example call: transcribe an audio file (batch job, wait for result): POST https://eu1.asr.api.speechmatics.com/v2/jobs/?wait=60&format=txt ↗ Unit price: Pro pay-as-you-go: batch Standard $0.24 per audio hour, batch Enhanced $0.40/hr, Melia 1 $0.129/hr; real-time Standard $0.24/hr, Enhanced $0.43/hr ↗ MCP tools: 4: transcribe_file, transcribe_directory, get_transcript, get_usage ↗	The free tier also includes 2 concurrent real-time sessions. No official vendor MCP server was found.	—

Speech and audio APIs and MCP servers for AI agents

Other categories