Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
VOCALSTACK_API_KEYNoYour VocalStack API key. Without a key the server still starts and lists its tools, but every tool call asks you to set one.
VOCALSTACK_MCP_URLNoAnother server address, for testing.https://api.vocalstack.com/v1/mcp

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

VocalStack transcribes and translates audio and video, recorded or live. Start a job with transcribe_url (a link) or transcribe_file (base64, up to 10 MiB); both return a transcriptionId at once. Follow it with wait_for_transcription (blocks up to 60 s; call again while timedOut is true), then read it with get_transcription. list_transcriptions and search_transcriptions find earlier work, live transcriptions included; search_by_meaning finds the passages that answer a question about it, in any language; get_summary and translate_transcription work on finished transcriptions. Broadcasts are live events listeners follow at a public link, reading or hearing what is said in their own language: list_broadcasts and get_broadcast show them (link, on air, listeners); start_broadcast transcribes and translates a live stream on one, get_transcription reads it as it runs, and stop_broadcast ends it.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}

Tools

Functions exposed to the LLM to take actions

NameDescription
transcribe_urlA

Start transcribing audio or video at a public URL (a file link, a podcast episode, a YouTube video, a Vimeo, TikTok, X or SoundCloud post…). Returns at once with a transcriptionId; call wait_for_transcription to follow it, then get_transcription for the text. Set summary for a title and summary, shareLink for a public link to the transcript.

transcribe_fileA

Start transcribing an audio or video file you have, sent as base64 (at most 10 MiB decoded). Returns at once with a transcriptionId; call wait_for_transcription next. Set summary for a title and summary, shareLink for a public link to the transcript.

get_transcriptionA

Get a transcription's status and its text with timestamps and speakers: the whole text once it is done, or the text so far while a file is transcribing or a live transcription is on air (partial is true). Use format "text" for plain readable text, "segments" for timed lines. Long transcripts are paged: ask again with nextOffset.

wait_for_transcriptionA

Wait for a file or link transcription to finish, up to 60 seconds per call. Returns its status; if timedOut is true it is still running, so call this again. When state is done, call get_transcription for the text. A live transcription runs until it is stopped, so this returns its status at once.

list_transcriptionsA

List this account's transcriptions, newest first, with their status: files and links (kind file) and live transcriptions, including those of broadcasts (kind live). Use search_transcriptions to find what was said.

search_transcriptionsB

Search what was said across this account's transcriptions, in any language. Returns matching lines with the transcription they are in and the second they were said.

search_by_meaningA

Find the passages of this account's transcriptions closest in meaning to a question or topic, in any language, even when they share none of its words (vector search). Returns up to 10 passages, closest first, with the transcription, the second it starts and how close it is (0 to 1). Use them to answer questions about what was said, and cite the transcription and second.

get_summaryA

Get the title and summary of a finished transcription, optionally in another language (when it was made with summaryTranslations). Asking for it writes one if the transcription was made without a summary. If state is pending the summary is being written; call again in a few seconds. not_requested means the recording is under a minute and is never summarised.

translate_transcriptionA

Get a finished transcription, from a file, a link or a live broadcast, translated into another language, as text or timed segments. Large transcriptions take up to a minute; translations are kept, so asking again is quick.

list_voicesA

List the voices that can speak a language: id, name, gender and accent when known, and a quality rating. For information only: VocalStack picks the voice that reads each speaker; the only choice is whether cloned voices are used.

list_speakersA

List the speakers this account saved: the people recognised by name in its transcriptions. Each has an id, a name, and whether their voice is cloned (cloned voices can read translations, see list_voices).

list_broadcastsC

List this account's broadcasts: live events listeners follow at a public link, reading (or hearing) what is said, translated into their own language as it is spoken; each listener picks from up to 10 languages, free, with no app or sign-up. For each: its link, whether it is on air, how many are listening, and the live transcription it is showing.

get_broadcastA

Get one broadcast: its public link, whether it is on air and since when, how many are listening, the live transcription it is showing (read it with get_transcription) and its earlier ones, and the streams it can record.

start_broadcastA

Start live transcription and translation of a live stream (a YouTube live, an internet radio station, an HLS or Icecast stream) on a broadcast. Returns the public link to share: anyone who opens it reads (or, with spokenTranslations, hears) the stream in their own language as it plays, and the transcriptionId to read with get_transcription. Billed per second of live transcription while it runs; listeners are free. It runs until stop_broadcast or the stream ends.

stop_broadcastA

Stop the live transcription a broadcast is running. The transcript is kept (get_transcription, translate_transcription, get_summary) and the link stays, for the next time it goes on air.

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A3.9/5.0

Scored across 15 tools

Disambiguation5/5

Each tool has a distinct resource+action target: transcribe_url vs transcribe_file differ by source, search_transcriptions (keyword) vs search_by_meaning (vector) are explicitly contrasted, and list_broadcasts/get_broadcast/start/stop_broadcast form clear boundaries. The only potentially confusable pair, list_voices vs list_speakers, is disambiguated by the descriptions (TTS catalogue vs saved people).

Naming Consistency5/5

All 15 names are snake_case and verb-first (list_*, get_*, transcribe_*, start_/stop_*, search_*, wait_for_*, translate_*). The only slight variation is search_by_meaning, which still reads naturally as a verb-led compound and fits the pattern.

Tool Count5/5

15 tools sit at the top of the reasonable band but each maps to a real capability of the domain (ingest file/URL, wait, fetch, search, summarize, translate, broadcast lifecycle, voice/speaker catalogues). Nothing looks redundant or padded.

Completeness3/5

Core flow (transcribe, wait, get, search, summarize, translate, broadcast start/stop) is well covered, but lifecycle endpoints are missing: there is no create_broadcast despite start_broadcast requiring an existing broadcast, no delete for transcriptions or broadcasts, and no create/rename/delete for speakers. These force the agent to dead-end or fall back to the web UI for setup and cleanup.

Maintenance

ActivityMaintained
ResponsivenessNo recent community issues