crispasr-agent-transcriber
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@crispasr-agent-transcriber transcribe meeting_recording.wav"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
# crispasr-agent-transcriber
Local-only transcription for Codex and MCP-based AI agents, powered by CrispASR. No cloud uploads, no API keys required for transcription.
GitHub Release | npm installer | PyPI package | MCP Registry
What it does
Give it a local audio or video file. It:
Probes the spoken language (English or Chinese) using CrispASR's FireRed LID.
Starts a local CrispASR server with the right backend -- Cohere Transcribe for English, Qwen3-ASR for Chinese.
Extracts audio from video with ffmpeg when needed.
Calls CrispASR's
/v1/audio/transcriptionsendpoint.Writes the transcript and metadata to disk.
For video understanding, captures synchronized keyframes and writes an agent-readable manifest.
Everything runs on your machine. Media never leaves it.
Related MCP server: whisper-transcribe-mcp
Quick install for Codex
The plugin includes the Codex Skill, command-line tool, and MCP server. Media
stays on your computer. Model files are never downloaded during install/update;
use the explicit models command when you want the installer to fetch them.
1. Install prerequisites
Install Node.js 20 or newer, uv, and
ffmpeg. The installer uses uv to provide Python.
node --version
uv --version
ffmpeg -version2. Run the installer
npx @emiyakatuz/crispasr-agent-transcriber@latest installThe installer:
downloads the matching GitHub Release and verifies its SHA-256 checksum;
installs the plugin under
~/plugins/crispasr-agent-transcriber;installs the Python and MCP dependencies;
detects CUDA, Vulkan, or CPU and installs the best CrispASR build;
registers the plugin in the Codex Personal marketplace;
preserves existing models, binaries, and outputs during updates.
3. Add the local models
Download the recommended local English, Chinese, and language-detection bundle:
npx @emiyakatuz/crispasr-agent-transcriber@latest modelsThe command downloads only approved GGUF files into:
~/plugins/crispasr-agent-transcriber/models/Then verify the installation:
npx @emiyakatuz/crispasr-agent-transcriber@latest doctor4. Enable the plugin
With a Codex build that supports plugin commands, run:
codex plugin add crispasr-agent-transcriber@personalIf the CLI has no codex plugin command, open the Codex desktop Plugins view
and install CrispASR Transcriber from the Personal marketplace. Start a new
conversation, then ask:
Transcribe C:\path\to\sample.mp4 with CrispASR using auto language detection.
Save a verbose JSON transcript and an SRT subtitle file.Update or uninstall
npx @emiyakatuz/crispasr-agent-transcriber@latest update
npx @emiyakatuz/crispasr-agent-transcriber@latest uninstallUninstall preserves local models, CrispASR binaries, and outputs. Use
uninstall --purge-data only when those files should also be deleted. See
Plugin installation for manual installation and
troubleshooting.
Direct command-line use
After installation, you can run the transcription script without Codex:
Set-Location (Join-Path $HOME "plugins\crispasr-agent-transcriber")
uv run python scripts/transcribe.py sample.mp4 --profile auto `
--manage-server `
--models-dir models `
--format verbose_jsonUse with other AI agents
The MCP server is the cross-agent interface. Any agent that supports MCP stdio can run the released package directly from GitHub:
uvx --from "crispasr-agent-transcriber[mcp] @ git+https://github.com/EmiyaKatuz/crispasr-agent-transcriber.git@v0.4.0" crispasr-agent-mcpUse the same command and arguments in Claude Desktop, Cursor, or another MCP client. See AI agent integrations for a generic MCP configuration and Codex CLI command.
Maintainer publishing
End users do not need the release steps. Maintainers should follow the publishing guide for Codex Marketplace, PyPI, MCP Registry, and cross-agent distribution.
Required models
Install/update never downloads models. Use the explicit models command or
download these three recommended GGUF files into a local directory such as
models/:
Purpose | Local file | Variant / size | Model page | File page |
English ASR |
| Q4_K, smaller default | ||
Chinese ASR |
| Q4_K, smaller default | ||
Language detection |
| Q4_K default |
Optional model IDs include english-q5-0, english-q5-1, english-q6,
english-q8, english-f16, chinese-q8, chinese-f16, lid-q2, lid-q8,
and lid-f16. Download a specific option with:
npx @emiyakatuz/crispasr-agent-transcriber@latest models --model-id english-q8All three upstream model families are Apache 2.0 licensed.
For automatic English/Chinese routing, pass both ASR paths. The language probe runs first, and only the matching model is loaded:
--english-model models\cohere-transcribe-q4_k.gguf
--chinese-model models\qwen3-asr-1.7b-q4_k.gguf
--lid-backend firered --lid-model models\firered-lid-q4_k.ggufFor an explicit english or chinese profile, --model remains available as
a single-model override.
CrispASR binary management
The tool auto-detects, installs, and updates the CrispASR binary from GitHub releases.
Flag | Effect |
| Download latest platform binary to |
| Upgrade to newest release |
| Show installed version + update availability |
| Custom directory (default |
| Exact path to |
When --manage-server is set and no binary is found, it auto-installs before
starting the server.
GPU detection
On install and update, the tool checks your hardware:
CUDA --
nvidia-smiavailable, orCUDA_PATH/CUDA_HOMEset, or CUDA inPATH-> downloadscrispasr-*-cudavariant.Vulkan --
vulkaninfoorVULKAN_SDKset (only when CUDA is absent) -> downloadscrispasr-*-vulkanvariant.CPU -- fallback when no GPU toolkit is detected.
macOS always uses the universal binary.
Profiles
Profile | Backend | ASR model | Language hint |
|
| Cohere Transcribe 03-2026 |
|
|
| Qwen3-ASR 1.7B |
|
| determined by LID | determined by LID | detected |
auto mode runs FireRed language detection on the media, then routes English
to Cohere or Chinese to Qwen3-1.7B. Mixed or uncertain content stops with a
clear error asking you to re-run with --profile english or --profile chinese.
Usage
Managed server (tool starts CrispASR for you)
uv run python scripts/transcribe.py sample.wav `
--profile auto `
--manage-server `
--models-dir models `
--format srt `
--out-dir outputsAdd --keep-server to leave the server running after transcription.
Manual server (you start CrispASR)
# Terminal 1 -- start the server
crispasr --server --backend cohere `
-m models\cohere-transcribe-q4_k.gguf `
--port 8080
# Terminal 2 -- transcribe
uv run python scripts/transcribe.py sample.mp4 `
--profile english `
--server-url http://127.0.0.1:8080 `
--format verbose_jsonIf the running server's backend doesn't match the selected profile, the tool prints the exact command you need to start the correct server.
Output formats
| File extension | Contents |
|
| Plain transcript |
|
| Full response with segments |
|
| SubRip subtitles |
|
| WebVTT subtitles |
A .metadata.json sidecar is always written alongside the transcript.
Video files
Video files are detected automatically. ffmpeg extracts the audio track to a temporary mono 16 kHz WAV before sending it to CrispASR. The temporary file is deleted when transcription finishes.
All CLI flags
--profile auto|english|chinese
--format text|verbose_json|srt|vtt|json
--out-dir PATH
--server-url URL
--allow-remote-server
--manage-server
--keep-server
--model PATH Local GGUF override for an explicit profile
--english-model PATH Cohere model selected after English detection
--chinese-model PATH Qwen3-ASR model selected after Chinese detection
--models-dir PATH Directory containing approved local GGUF models
--allow-model-auto-download
--lid-model PATH Local LID model path
--lid-backend firered|silero|ecapa|whisper
--host HOST Managed server host (default 127.0.0.1)
--port PORT Managed server port (default 8080)
--language CODE Language hint for transcription
--prompt TEXT Initial prompt/context
--vad Enable voice activity detection
--diarize Enable speaker diarization
--diarize-method METHOD
--hotwords WORD,WORD Comma-separated hotwords
--no-timestamps
--preprocess auto|always|never
--api-key KEY If CRISPASR_API_KEYS is enabled
--crispasr-bin-dir PATH
--crispasr-bin PATH
--install-crispasr
--update-crispasr
--crispasr-status
--list-models
--download-models
--model-id MODEL_ID
--overwrite-modelsMCP server
uv sync --extra mcp
uv run --extra mcp crispasr-agent-mcpExposed tools:
Tool | Description |
| Check CrispASR server health |
| List available backends |
| Run language detection on a file |
| List approved model choices and local install status |
| Explicitly download approved model files |
| Return recommended local model paths |
| Transcribe an audio file |
| Transcribe a video file |
| Transcribe a video, capture synced keyframes, and return an agent context |
| Batch-transcribe a folder |
Security model
No cloud uploads. Media files stay on the local filesystem.
No remote servers by default.
--server-urlonly accepts localhost unless--allow-remote-serveris explicitly passed.No URL inputs. Only local file paths are accepted. URLs, S3, and other remote schemes are rejected.
No shell injection. ffmpeg is called with argument lists and
shell=False. No user-controlled strings are interpolated into shell commands.No implicit model downloads. Install/update never downloads models, and CrispASR model auto-download (
-m auto) requires--allow-model-auto-download. Themodelscommand andcrispasr_download_modelstool download only allowlisted Hugging Face files.Temporary files are cleaned up. Converted WAV files and LID probe windows are deleted when transcription finishes.
Binary downloads are explicit. CrispASR binary installs only from the official
CrispStrobe/CrispASRGitHub releases.Verified plugin releases. The npm installer requires the plugin ZIP to match the SHA-256 value published in the same GitHub Release.
Narrow installer writes. The installer manages only its plugin directory and the named Personal marketplace entry. Updates preserve local models, binaries, and outputs.
Generated understanding stays local. Video keyframes, manifests, and agent context files are written under the selected output directory and are ignored by Git.
Verify
uv run pytest
uv run ruff check . # zero lint warningsLicense
This project is licensed under the MIT License.
Third-party components and attribution
This tool orchestrates several independently-licensed projects. It does not bundle, fork, or redistribute their code -- it downloads pre-built binaries and calls them as subprocesses or HTTP services at runtime.
Component | License | Role |
MIT | ASR engine, server, language detection | |
LGPL 2.1+ / GPL 2+ | Media decoding and audio extraction | |
Apache 2.0 | English ASR model (loaded by CrispASR) | |
Apache 2.0 | Chinese ASR model (loaded by CrispASR) | |
Apache 2.0 | Language detection model (loaded by CrispASR) | |
BSD | HTTP client for CrispASR API | |
MIT | MCP server framework | |
MIT | npm installer runtime | |
MIT | Verified plugin ZIP extraction |
Model files must be downloaded separately by the user from their respective HuggingFace repositories. See Required models above.
Related projects
CrispASR -- the ASR engine this tool wraps
CrisperWeaver -- CrispASR's desktop GUI (not used by this tool)
Available Tools
10 toolscrispasr_backendsC
List available backends from a running CrispASR server.
Inputs: server_url.
crispasr_detect_languageA
Run language detection on a media file using CrispASR LID.
Returns the detected language and routing decision (english / chinese / uncertain). Requires a local LID model path. The firered backend is recommended.
Inputs: file_path (required), lid_model (required), lid_backend, crispasr_bin.
crispasr_download_modelsA
Download approved GGUF models into a local models directory.
Downloads only from the built-in allowlist. By default, installs the recommended English, Chinese, and language-detection models.
Inputs: model_ids, overwrite, models_dir.
crispasr_healthB
Check whether a CrispASR server is running and what backend it uses.
Inputs: server_url.
crispasr_list_modelsA
List approved local GGUF model choices and installation status.
Inputs: models_dir.
crispasr_resolve_model_pathsC
Return recommended local model paths for auto-routed transcription.
Inputs: models_dir.
transcribe_audioA
Transcribe a local audio file through a CrispASR server.
Supports auto language routing (needs --lid-model) or explicit english / chinese profiles. Can start a managed server on demand.
Inputs: file_path (required), model, out_dir, profile, +10 more.
transcribe_folderA
Batch-transcribe all supported media files in a folder.
Keeps a managed server running across all files when --manage-server and --keep-server are both set.
Inputs: folder_path (required), model, out_dir, profile, +10 more.
transcribe_videoA
Transcribe a local video file through a CrispASR server.
Extracts audio with ffmpeg before sending to CrispASR. Supports the same options as transcribe_audio.
Inputs: file_path (required), model, out_dir, profile, +10 more.
understand_videoC
Transcribe a local video, capture synced keyframes, and persist an agent context.
Inputs: file_path (required), model, out_dir, profile, +14 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 10 tools
Most tools are clearly separated by domain: server health/backends, model management, language detection, and transcription tasks. The main ambiguity is between transcribe_video and understand_video, since both transcribe local video, and crispasr_backends vs crispasr_health both report server state.
The set is readable and consistently snake_case, but conventions are mixed: crispasr_backends and crispasr_health are noun-style, crispasr_* management tools use verb_noun, transcribe_* forms its own group, and understand_video is a vague outlier.
10 tools is well within the ideal range and each area of the domain—server health, backends, models, language detection, audio/video transcription, batch processing, and agent context—is represented. The size fits the server's scope without feeling bloated.
The core workflow is covered end-to-end: check the server, download models, resolve paths, detect language, and transcribe individual or batch media files. Minor gaps exist, such as no model removal tool and no explicit management/retrieval for the agent context persisted by understand_video.
Maintenance
Related MCP Connectors
MCP server for Speech-to-Text
MCP server for RiverScript, an AI transcription platform - fetches transcripts shared via a link.
An MCP server that gives any LLM or agent clean YouTube transcripts on demand: a single video, a whole channel, or a playlist, plus AI cleanup of auto-generated captions. API-key auth, credit-based, same backend as the public v1 API. Get a free API key with 25 free credits at youtubetranscriptdownload.com/account.
Hong Kong Cantonese & Konglish AI voiceover + frame-synced SRT/VTT subtitle MCP server.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceMCP server for offline speech-to-text and speaker diarization, enabling AI agents to transcribe audio locally without cloud APIs.86 PyPI4MIT
- AlicenseAqualityBmaintenanceMCP server for audio transcription using local faster-whisper or OpenAI Whisper API, enabling multilingual transcription with optional GPT post-processing.3160 PyPIMIT
- AlicenseAqualityDmaintenanceLocal, private audio transcription MCP server enabling AI agents to transcribe audio files entirely on-device without uploading data.3MIT

FunASR MCP Serverofficial
AlicenseNot gradedqualityAmaintenanceLocal speech recognition MCP server powered by FunASR and SenseVoice. It provides privacy-friendly audio transcription through a transcribe_audio tool.21,048MIT