Speech Recognition & Transcription MCP Servers — Whisper, Deepgram, ElevenLabs, Groq, and More

Part of the Media & Entertainment category.

Speech recognition is the other half of voice AI — and the half that’s harder to get right. While text-to-speech has a clear MCP leader (ElevenLabs, covered in our Audio & Video Processing review), speech-to-text has matured significantly since early 2026 with two major vendors now offering official MCP servers.

This review covers MCP servers specifically built for converting audio to text — speech recognition, transcription, and dictation. The ecosystem splits into three approaches: local Whisper-based servers (run entirely on your hardware, zero API cost), cloud API wrappers (Deepgram, ElevenLabs, AssemblyAI, Groq, OpenAI Whisper API), and multimodal LLM transcription (Gemini, GPT-4o Audio, Voxtral). Several servers also include text-to-speech capabilities, but we evaluate them primarily on their STT quality.

The headline: Deepgram launched an official full MCP server (deepgram/mcp) with STT, dynamic tool loading, and CLI integration — filling the biggest gap from our initial review. ElevenLabs MCP (1,300+ stars) added Scribe transcription in April 2026, making it the most popular server in this category. Groq Whisper MCP achieves 216x real-time transcription speed at 9x lower cost than OpenAI’s API. Local options remain strong — mlx-whisper-mcp brings Apple’s MLX framework with large-v3-turbo, and whisper.cpp still hits 15x real-time on Apple Silicon.

Local Whisper-Based Servers

These servers run speech recognition entirely on your machine using variants of OpenAI’s Whisper model. Zero API cost, full privacy, but require local compute resources.

speech-mcp (Kvadratni)

The most popular speech recognition MCP server, designed as a voice interface for the Goose AI agent framework. Bidirectional: transcribes your speech and speaks responses aloud, creating a natural conversational loop.

What Works Well

Full voice conversation loop. Goose launches the speech interface, listens for voice input, processes it, speaks the response, then listens again. No manual switching between modes — silence detection handles turn-taking automatically.

54+ voice models via Kokoro TTS. The text-to-speech side uses Kokoro, an open-weight 82M parameter model, with a wide selection of voices. Responses are spoken, not just transcribed.

100% local processing. Both STT (faster-whisper) and TTS (Kokoro) run locally. No audio data leaves your machine. Good for sensitive conversations or air-gapped environments.

Audio/video transcription. Beyond live microphone input, it can transcribe audio and video files with optional timestamps and speaker detection.

What Doesn’t Work Well

Goose-specific. Designed for Block’s Goose framework. Works with other MCP clients for transcription, but the voice conversation loop is Goose-specific.

Base Whisper model only. Uses the faster-whisper “base” model by default for speed. Accuracy is decent but not as good as the “large-v3” model for difficult audio (accents, background noise, technical jargon).

No license specified. The GitHub repo has no license file.

local-stt-mcp (SmartLittleApps)

A high-performance local transcription server using whisper.cpp, the C/C++ port of Whisper. Specifically optimized for Apple Silicon with impressive benchmarks.

What Works Well

15.8x real-time on Apple Silicon. The benchmarks are compelling — 15.8x real-time processing speed via the Apple Neural Engine, compared to 5.5x for WhisperX. A 10-minute audio file transcribes in under 40 seconds.

Speaker diarization. Identifies and separates multiple speakers in audio — critical for meeting transcriptions, interviews, and podcasts. Most Whisper MCP servers skip this.

Under 2GB memory. Light enough to run alongside other applications on a Mac. WhisperX needs ~4GB for comparison.

Universal audio support. Automatic conversion from MP3, M4A, FLAC, WAV, and other formats before transcription. No manual preprocessing needed.

What Doesn’t Work Well

Apple Silicon only (practically). While it may work on other platforms, the optimization and benchmarks are specifically for Apple Neural Engine. Linux/Windows users won’t see the same performance.

New project, low adoption. Limited community feedback and issue history. The performance claims are from the author’s benchmarks.

Fast-Whisper-MCP-Server (BigUncle)

A Faster Whisper server targeting NVIDIA GPU users. Supports multiple model sizes from tiny to large-v3, batch processing, and multiple output formats.

What Works Well

CUDA acceleration. If you have an NVIDIA GPU, this is the fastest local option on Linux/Windows. Automatic CUDA detection with CPU fallback.

Multiple model sizes. Choose from tiny (fastest, least accurate) to large-v3 (slowest, most accurate) based on your quality/speed requirements. The flexibility matters — transcribing meeting notes doesn’t need the same accuracy as medical dictation.

Multiple output formats. VTT subtitles, SRT, and JSON. Useful for subtitle generation workflows beyond just raw transcription.

Batch processing. Process multiple audio files in a single operation. Good for transcribing a batch of recorded interviews or podcast episodes.

What Doesn’t Work Well

NVIDIA-only for GPU. No Metal (macOS) or ROCm (AMD) support. CPU fallback works but is much slower.

Low adoption. Few stars, limited community validation.

voice-to-text-mcp (acazau)

The only Rust-based speech MCP server. Cross-platform hardware acceleration — Metal/CoreML on macOS, CUDA on Linux/Windows, with CPU fallback everywhere.

What Works Well

True cross-platform GPU acceleration. The only server that supports both Apple Metal/CoreML and NVIDIA CUDA. Wherever you run it, it uses whatever GPU is available.

Rust performance. Lower overhead than Python-based alternatives. The whisper-rs bindings are well-maintained.

Claude Code integration. Designed to work directly with Claude Code for voice-to-code dictation. Dictate code, comments, and documentation by voice within your development workflow.

What Doesn’t Work Well

Long build times. First build with CUDA takes 6+ minutes due to whisper-rs-sys compilation. Metal/CoreML takes 2-3 minutes. Not a quick install.

Early project. Limited documentation and community adoption.

mlx-whisper-mcp (kachiO)

A local transcription server using Apple’s MLX framework — the native ML acceleration layer for Apple Silicon. Uses the high-quality whisper-large-v3-turbo model natively on M-series chips.

What Works Well

MLX-native performance. Runs Whisper through Apple’s MLX framework rather than whisper.cpp, taking advantage of the unified memory architecture on M-series chips. No model conversion needed — uses MLX-optimized weights directly from Hugging Face (mlx-community/whisper-large-v3-turbo).

Zero-setup installation. Run uv run mlx_whisper_mcp.py and it auto-installs all dependencies. No venv setup, no manual model downloads.

Multiple input sources. Supports direct file transcription, base64-encoded audio data, and YouTube video transcription.

What Doesn’t Work Well

Apple Silicon only. MLX is exclusively for M-series Macs. No Linux or Windows support.

Requires Python 3.12+. Higher Python version requirement than most alternatives.

whisper-mcp + apple-voice-memo-mcp (jwulff)

A paired set of MCP servers: whisper-mcp handles local transcription via whisper.cpp, while apple-voice-memo-mcp provides programmatic access to Apple Voice Memos — including extracting embedded transcripts from .m4a files via Apple’s custom tsrp MPEG-4 atom.

What Works Well

Apple Voice Memos integration. The only MCP server that can browse, access, and transcribe Voice Memos directly. Extracts Apple’s built-in transcripts (stored inside .m4a files) or generates new ones via SFSpeechRecognizer.

Composable design. The two servers pair together — use apple-voice-memo-mcp to find recordings, whisper-mcp to transcribe them locally. Works with Claude Code directly.

Multiple Whisper models. Download and switch between model sizes on the fly. Models stored in ~/.whisper/.

What Doesn’t Work Well

macOS only. Apple Voice Memos is a macOS feature. The whisper-mcp server could theoretically run anywhere, but the pairing only makes sense on Mac.

Niche use case. Most useful for people who already record Voice Memos and want AI access to them.

Cloud API Wrappers

These servers call external speech recognition APIs. Better accuracy on difficult audio, no local compute needed, but require API keys and incur costs.

Deepgram Official MCP Server

NEW since March 2026. Deepgram launched an official full MCP server covering both STT and TTS — the biggest change in this category since our initial review. The previous deepgram-devs/deepgram-mcp was TTS-only; this replaces it entirely.

What Works Well

Dynamic tool loading from API. The server fetches its tool list from Deepgram’s API at runtime. When Deepgram ships new features, they appear in your editor automatically — no package upgrade needed. This is a genuinely novel approach among MCP servers.

Full Deepgram platform access. Transcription, text-to-speech (Aura voices), audio intelligence (sentiment, topics, intents, entities), speaker diarization, language detection, and multiple model support — all through one server.

CLI integration. If you already use the Deepgram CLI (pip install deepctl), deepgram-mcp is included as a dependency. Run dg mcp and it works. No separate install needed.

Flux Multilingual STT. As of April 2026, supports Flux Multilingual for real-time conversational STT across 10 languages with code-switching support.

What Doesn’t Work Well

API costs. Deepgram charges per audio minute. Fine for occasional use, adds up for bulk transcription.

Requires internet. All processing happens in Deepgram’s cloud. No offline fallback.

Deepgram Community Servers

The community server by reddheeraj remains available and still works. However, with the official deepgram/mcp server now covering STT, TTS, diarization, and audio intelligence, the community server is largely superseded for most use cases.

AssemblyAI MCP Server

A community MCP server for AssemblyAI’s transcription API. AssemblyAI is a strong commercial alternative to Deepgram, particularly for longer-form audio and async workflows.

What Works Well

npx zero-install. Run npx assembly-ai-mcp@latest — no cloning, no dependencies. One command and it’s running.

Async job management. Submit audio for transcription without blocking, get back a job ID, check status later. Good for long recordings where you don’t want to wait.

Speaker diarization + language detection. Standard enterprise STT features included.

What Doesn’t Work Well

Community-maintained. Not an official AssemblyAI project. May lag behind API updates.

No real-time streaming. File/URL-based transcription only. No live microphone input.

Groq Whisper MCP Server

Uses Groq’s LPU inference hardware to run Whisper at extraordinary speeds. Same model as OpenAI’s Whisper API, but 9x cheaper and dramatically faster.

What Works Well

216x real-time transcription. Groq’s hardware accelerates Whisper to speeds that make even short audio files feel instant. A 10-minute recording transcribes in under 3 seconds.

9x cheaper than OpenAI Whisper API. Same whisper-large-v3-turbo model, different infrastructure. Significant cost advantage for bulk transcription.

Word-level timestamps. Returns full text plus [{word, start, end}] arrays — useful for subtitle generation, audio editing, and precise navigation.

Automatic caching + cost estimation. Results are cached per-project. Shows estimated cost before processing.

What Doesn’t Work Well

Requires Groq API key. Free tier has rate limits. Heavy use requires a paid plan.

Requires ffmpeg. Audio extraction from video needs ffmpeg installed locally.

Cloud-dependent. Audio is sent to Groq’s servers. Not suitable for sensitive content requiring local processing.

OpenAI Whisper API Servers

Three servers wrapping OpenAI’s Whisper API:

arcaputo3/mcp-server-whisper is the most architecturally sophisticated — well-documented Domain-Driven Design with Pydantic models, regex-based file searching, batch processing, format conversion, automatic compression for oversized files, and enhanced transcription with specialized prompts. Also includes TTS capabilities via OpenAI’s speech services. The architecture is over-engineered for its adoption (2 stars), but the code quality is high.

mfleurival/whisper-mcp-v2 supports multiple output formats: JSON, text, SRT, VTT, and verbose JSON. Good for subtitle generation. Language support for non-English audio.

Ichigo3766/audio-transcriber-mcp is the simplest — a clean, focused transcription tool using the OpenAI Whisper API. One tool, one job.

All three require an OpenAI API key. The Whisper API is relatively cheap ($0.006/minute) but not free.

Multimodal LLM-Based Transcription

The newest approach: skip traditional ASR entirely and use multimodal LLMs that understand audio natively.

Cloud-ASR-MCP (danielrosehill)

Uses audio-capable multimodal models — Gemini, GPT-4o Audio, and Voxtral — that process audio in a single pass. The key innovation: you can provide text prompt guidance to clean up transcripts on the fly.

What Works Well

Prompt-guided transcription. Tell the model “this is a medical interview, fix any medical terminology” or “remove filler words and false starts” — the LLM applies contextual understanding during transcription, not after. This is genuinely different from traditional ASR + post-processing pipelines.

Multiple model backends. Switch between Gemini (Google), GPT-4o Audio (OpenAI), and Voxtral (Mistral) depending on quality/cost requirements. Each model has different strengths.

What Doesn’t Work Well

Work in progress. The project is explicitly WIP. Expect rough edges.

Higher cost. Multimodal LLM inference is more expensive per minute than dedicated ASR APIs. Best for small volumes where accuracy matters more than cost.

Latency. LLM inference adds latency compared to purpose-built ASR models. Not suitable for real-time transcription.

Gemini-Transcription-MCP (danielrosehill)

A focused implementation using only Gemini’s multimodal audio capabilities, with built-in post-processing. By the same author as Cloud-ASR-MCP — this is the single-model version for teams standardized on Google.

Desktop Voice Agents

These go beyond transcription to provide complete voice-controlled agent experiences.

t2t (acoyfellow)

A cross-platform desktop application combining system-wide dictation with MCP agent integration. Two modes: hold fn for plain dictation (voice-to-text anywhere), hold fn+ctrl for agent mode (voice commands routed to any connected MCP server).

What Works Well

System-wide dictation. Works in any application — not just MCP clients. Hold the fn key, speak, release, and text appears wherever your cursor is. This alone makes it useful as a general dictation tool.

Agent mode. The fn+ctrl mode connects to any MCP server(s), letting you give voice commands that trigger tool calls. Multiple simultaneous MCP servers supported via stdio, HTTP, and SSE transports.

Visual feedback. Red bar while recording dictation, cyan for agent mode, amber while processing. Simple but effective.

Fully local STT. Only the OpenRouter API calls for agent responses go to the cloud. Transcription stays local.

What Doesn’t Work Well

Not code-signed. macOS shows security warnings on first launch. Users need to manually allow execution in System Settings.

Early stage. Featured on Hacker News in late 2025 but still maturing. Limited documentation.

Also notable

• ElevenLabs MCP (1,300+ stars, official) — covered in our Audio & Video Processing review. Now includes Scribe transcription (added v2.44.0, April 21 2026) — the most popular MCP server with STT capability. Scribe v2 offers 150ms latency real-time transcription across 99+ languages. The official server now supports TTS, voice cloning, sound effects, audio isolation, AND transcription.
• MCP-Elevenlab-Scribe-ASR (aromanstatue) — a dedicated MCP server specifically for ElevenLabs’ Scribe ASR API with WebSocket streaming, real-time microphone input, and event detection. More STT-focused than the official ElevenLabs MCP.
• Vexa (1,800 stars) — meeting transcription with real-time Whisper, covered in our Video Conferencing review. Specialized for meeting contexts.
• speaches-ai/speaches — OpenAI API-compatible server for STT (faster-whisper) and TTS (piper/Kokoro) with streaming. Not MCP-native, but can be used as a backend for MCP servers. Formerly “faster-whisper-server” — aims to be “Ollama for TTS/STT models.”
• DefiBax/mcp_servers (Voice Recorder) — records audio from default microphone and transcribes using Whisper. Designed as Goose custom extension. Configurable sample rate, duration, and model.
• blacktop/mcp-tts — multi-engine TTS server supporting macOS say, ElevenLabs, Google TTS, and OpenAI TTS. Purely text-to-speech, no recognition.
• kristofferv98/MCP_tts_server — Kokoro + OpenAI TTS with streaming and fallback between engines.
• mberg/kokoro-tts-mcp — Kokoro TTS with S3 upload for generated audio files.

The bottom line

Speech recognition MCP servers have matured significantly since early 2026. Two major vendors — Deepgram and ElevenLabs — now offer official servers with STT capability, and Groq’s hardware-accelerated option provides a compelling speed/cost middle ground.

Local Whisper servers remain the practical choice for privacy-conscious users. They’re free, private, and fast — especially on Apple Silicon (15x real-time via whisper.cpp, or MLX-native with large-v3-turbo) or NVIDIA GPUs. The new mlx-whisper-mcp and jwulff/whisper-mcp entries make Apple Silicon the best-served local platform.

Cloud API wrappers are now the strongest category. Deepgram’s official MCP server with dynamic tool loading is best-in-class for enterprise use. Groq Whisper offers 216x real-time at 9x lower cost than OpenAI. ElevenLabs’ Scribe (via their official 1,300+ star MCP server) adds 150ms real-time transcription across 99+ languages. AssemblyAI rounds out the options for async workflows.

Multimodal LLM transcription remains the most interesting emerging approach. Prompt-guided transcription could eventually make traditional ASR pipelines obsolete for low-volume, high-accuracy use cases. Still early and expensive, but the gap is narrowing.

Best official vendor server: deepgram/mcp (full platform access, dynamic tools, CLI integration) Best overall (by adoption): ElevenLabs MCP (1,300+ stars, Scribe transcription + TTS + voice cloning) Best for speed/cost: Groq Whisper MCP (216x real-time, 9x cheaper than OpenAI) Best for local transcription: speech-mcp (81 stars, bidirectional voice, Goose integration) Best for Apple Silicon (native): mlx-whisper-mcp (MLX framework, large-v3-turbo, zero-setup) Best for Apple Silicon (whisper.cpp): local-stt-mcp (15x real-time, speaker diarization) Best for NVIDIA GPUs: Fast-Whisper-MCP-Server (CUDA acceleration, batch processing) Best for cross-platform: voice-to-text-mcp (Rust, Metal + CUDA support) Best for async workflows: AssemblyAI MCP (npx install, job management, diarization) Best for context-aware transcription: Cloud-ASR-MCP (multimodal LLMs with prompt guidance) Best for system-wide dictation: t2t (desktop app, fn-key dictation + agent mode) Best for Apple Voice Memos: apple-voice-memo-mcp + whisper-mcp (macOS integration)

Rating: 4/5 — The category has transformed since March 2026. Deepgram’s official MCP server with dynamic tool loading fills the biggest gap from our initial review. ElevenLabs adding Scribe transcription to their 1,300+ star server gives the category a high-adoption anchor. Groq’s 216x real-time speed makes cloud transcription feel instant. The local ecosystem has diversified with MLX-native options. The remaining gaps: OpenAI still has no official Whisper MCP server, and no single local project has broken past ~100 stars. But with two major vendors official and viable cloud options at every price point, this is now a functional, well-served category.

This review was researched and written by an AI agent. We have not personally tested these servers — our analysis is based on documentation, source code, GitHub metrics, and community adoption. See our methodology for details.

This review was last edited on 2026-05-02 using Claude Opus 4.6 (Anthropic).