Skip to main content
Glama
Venice MCP Server

Venice MCP Server

Official
by veniceai

Venice ASR (Speech-to-Text)

venice_asr

Transcribes audio from an http(s) or base64 data URL by fetching it server-side, returning JSON or text with optional timestamps and paging for long transcripts.

Instructions

Transcribe audio. Fetches the URL server-side and forwards as multipart/form-data file upload. Upstream transcription JSON larger than 1 MiB is rejected so word/character timestamps cannot exhaust memory. Timestamp arrays and transcripts longer than 8000 characters are paged in the MCP result: such a transcription returns result_handle alongside the first page, and every later page must be requested with that handle so the clip is transcribed — and charged — exactly once. Handles are held in memory, per MCP session, for 10 minutes and only the 16 most recent survive, after which paging fails and a fresh transcription is needed. Supports x402 wallet auth (no Venice account needed) and API key.

Tool call template

This tool declares no required parameters. Add optional parameters from the input schema as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {},
    "name": "venice_asr"
  }
}

Input Schema

NameRequiredDescriptionDefault
modelNo
languageNo
audio_urlNoAudio to transcribe, as an http(s) URL or an inline base64 data URL. Include a concrete media type in data URLs, e.g. audio/mpeg for MP3 or audio/mp4 for M4A. Required unless result_handle is supplied.
timestampsNoInclude word and character timestamps in JSON responses. Defaults to false.
text_offsetNoStart character of the transcript page (8000 characters each). Defaults to 0. Use result_handle with next_text_offset to read past the first page.
result_handleNoHandle returned by an earlier paged transcription. Pages that retained result without submitting another transcription, so audio_url is ignored when this is set. Expires after 10 minutes or once 16 newer results are retained; an expired handle is an error rather than a silent re-transcription.
response_formatNo
timestamp_limitNoMax entries returned per timestamp array. Defaults to 200. Follow next_timestamp_offset; it is null on the last page.
timestamp_offsetNoStart index into each timestamp array (word/segment/char). Defaults to 0. Use result_handle to move past the first page.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changedv0.3.1
    • changedInput schema / properties / audio_url / description
      Previous value: -"Audio to transcribe. Required unless result_handle is supplied."New value: +"Audio to transcribe, as an http(s) URL or an inline base64 data URL. Include a concrete media type in data URLs, e.g. audio/mpeg for MP3 or audio/mp4 for M4A. Required unless result_handle is supplied."
  2. Changed8 schema fields changedv0.3.0
    • addedInput schema / properties / audio_url / description
      Added value: +"Audio to transcribe. Required unless result_handle is supplied."
    • changedInput schema / properties / response_format / enum
      Previous value: -[
      -  "json",
      -  "text",
      -  "srt",
      -  "verbose_json",
      -  "vtt"
      -]New value: +[
      +  "json",
      +  "text"
      +]
    • addedInput schema / properties / result_handle
      Added value: +{
      +  "description": "Handle returned by an earlier paged transcription. Pages that retained result without submitting another transcription, so audio_url is ignored when this is set. Expires after 10 minutes or once 16 newer results are retained; an expired handle is an error rather than a silent re-transcription.",
      +  "minLength": 1,
      +  "type": "string"
      +}
    • addedInput schema / properties / text_offset
      Added value: +{
      +  "description": "Start character of the transcript page (8000 characters each). Defaults to 0. Use result_handle with next_text_offset to read past the first page.",
      +  "minimum": 0,
      +  "type": "integer"
      +}
    • addedInput schema / properties / timestamp_limit
      Added value: +{
      +  "description": "Max entries returned per timestamp array. Defaults to 200. Follow next_timestamp_offset; it is null on the last page.",
      +  "maximum": 500,
      +  "minimum": 1,
      +  "type": "integer"
      +}
    • addedInput schema / properties / timestamp_offset
      Added value: +{
      +  "description": "Start index into each timestamp array (word/segment/char). Defaults to 0. Use result_handle to move past the first page.",
      +  "minimum": 0,
      +  "type": "integer"
      +}
    • addedInput schema / properties / timestamps
      Added value: +{
      +  "description": "Include word and character timestamps in JSON responses. Defaults to false.",
      +  "type": "boolean"
      +}
    • removedInput schema / required
      Removed value: -[
      -  "audio_url"
      -]
  3. First observedv0.2.0

TDQS

A4.1/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Goes well beyond annotations, which only declare non-read-only, open-world, non-idempotent, non-destructive. The description discloses the 1 MiB upstream rejection and its memory rationale, page size (8000 chars, timestamp_limit default 200), in-memory per-session handles with a 10-minute / 16-result lifetime, exactly-once billing, and both x402 and API-key auth paths.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the core action, then layered into transport, limits, paging/handle semantics, and auth. Dense but every sentence carries required information; no filler.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 9-parameter tool with no output schema, the description covers the hard parts: paging protocol, handle lifetime, billing semantics, and auth. Remaining gaps are return-value shape details and defaults for model/language/response_format, which are minor given the paging contract is explicit.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

At 67% schema coverage the description adds real meaning for the paging parameters and the audio_url/result_handle interaction. However, model, language, and response_format have no description in either schema or prose, and the description does not compensate for those gaps.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose4/5

Does the description clearly state what the tool does and how it differs from similar tools?

Opens with a specific verb+resource ("Transcribe audio") and immediately clarifies the transport (server-side URL fetch, multipart/form-data upload). This separates it from sibling venice_tts by direction, though it never names a sibling explicitly.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives solid operational when-to-use guidance: when result_handle is supplied, audio_url is ignored; later pages must be requested with the handle; an expired handle is an error rather than a silent re-transcription. It stops short of naming alternatives (e.g., venice_tts, venice_audio_quote) or when not to use it.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.