Skip to main content
Glama
Swih

mistral-mcp

by Swih
OverviewSchemaRelated ServersScoreDiscussions
TypeScript
Remote

Voxtral speech-to-text

voxtral_transcribe
Read-onlyIdempotent

Transcribe audio files to text using Mistral Voxtral models. Accepts public URLs or uploaded files, with options for language hint, speaker diarization, and context bias.

Instructions

Transcribe an audio file to text using Mistral Voxtral.

Accepted models:

  • voxtral-mini-latest

  • voxtral-small-latest

Audio source is one of:

  • { type: "file_url", fileUrl: "https://..." } (public URL)

  • { type: "file", fileId: "" }

Options:

  • language: ISO-639-1 hint (e.g. 'fr', 'en'). Boosts accuracy when known.

  • temperature: sampling temperature.

  • diarize: return per-speaker segments (default false).

  • timestampGranularities: ['segment'] to return per-segment timestamps.

  • contextBias: list of phrases/terms that should bias the decoder.

Returns plain text, detected language, optional segments[], and token usage.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
audioYes
modelNoSTT model. Default: voxtral-mini-latest.
languageNoISO-639-1 language hint (e.g. 'fr', 'en').
temperatureNo
diarizeNo
timestampGranularitiesNoOnly 'segment' is currently supported.
contextBiasNo

Output Schema

TableJSON Schema
NameRequiredDescriptionDefault
textYes
modelYes
languageYes
segmentsNo
usageNo

Implementation Reference

  • src/tools-audio.ts:145-195 (handler)
    Handler function for voxtral_transcribe. Calls mistral.audio.transcriptions.complete with the audio source (fileUrl or fileId) and optional parameters (language, temperature, diarize, timestampGranularities, contextBias). Maps the response segments and returns text content plus structured metadata.
      async (input) => {
    try {
      const model = input.model ?? DEFAULT_STT_MODEL;
      const request: {
        model: string;
        fileUrl?: string;
        fileId?: string;
        language?: string;
        temperature?: number;
        diarize?: boolean;
        timestampGranularities?: Array<"segment">;
        contextBias?: string[];
      } = { model };
      if (input.audio.type === "file_url") request.fileUrl = input.audio.fileUrl;
      if (input.audio.type === "file") request.fileId = input.audio.fileId;
      if (input.language !== undefined) request.language = input.language;
      if (input.temperature !== undefined) request.temperature = input.temperature;
      if (input.diarize !== undefined) request.diarize = input.diarize;
      if (input.timestampGranularities !== undefined)
        request.timestampGranularities = input.timestampGranularities;
      if (input.contextBias !== undefined) request.contextBias = input.contextBias;

      const res = await mistral.audio.transcriptions.complete(
        request as never
      );

      const segments = res.segments?.map((s) => ({
        text: s.text,
        start: s.start,
        end: s.end,
        score: s.score ?? undefined,
        speaker_id: s.speakerId ?? undefined,
      }));

      const structured = {
        text: res.text,
        model: res.model,
        language: res.language,
        segments,
        usage: mapUsage(res.usage),
      };

      return {
        content: [toTextBlock(res.text)],
        structuredContent: structured,
      };
    } catch (err) {
      return errorResult("voxtral_transcribe", err);
    }
  }
);
  • src/tools-audio.ts:30-45 (schema)
    Output schema for voxtral_transcribe. Defines the TranscribeOutputShape with text, model, language, optional segments (each with text/start/end/score/speaker_id), and optional usage.
    const TranscriptionSegmentSchema = z.object({
  text: z.string(),
  start: z.number(),
  end: z.number(),
  score: z.number().optional(),
  speaker_id: z.string().optional(),
});

export const TranscribeOutputShape = {
  text: z.string(),
  model: z.string(),
  language: z.string().nullable(),
  segments: z.array(TranscriptionSegmentSchema).optional(),
  usage: UsageSchema.optional(),
};
export const TranscribeOutputSchema = z.object(TranscribeOutputShape);
  • src/tools-audio.ts:69-84 (schema)
    Input schema for audio source (file_url or file discriminator). Used as part of the voxtral_transcribe input.
    const AudioSourceSchema = z.union([
  z.object({
    type: z.literal("file_url"),
    fileUrl: z
      .string()
      .describe("HTTPS URL to an audio file (mp3/wav/flac/ogg/webm/m4a)."),
  }),
  z.object({
    type: z.literal("file"),
    fileId: z
      .string()
      .describe(
        "ID of an audio file previously uploaded via the Files API (purpose=audio)."
      ),
  }),
]);
  • src/tools-audio.ts:119-135 (schema)
    Input schema for voxtral_transcribe tool: audio source, optional model, language, temperature, diarize, timestampGranularities, and contextBias.
    inputSchema: {
  audio: AudioSourceSchema,
  model: SttModelSchema.optional().describe(
    `STT model. Default: ${DEFAULT_STT_MODEL}.`
  ),
  language: z
    .string()
    .optional()
    .describe("ISO-639-1 language hint (e.g. 'fr', 'en')."),
  temperature: z.number().min(0).max(2).optional(),
  diarize: z.boolean().optional(),
  timestampGranularities: z
    .array(z.enum(["segment"]))
    .optional()
    .describe("Only 'segment' is currently supported."),
  contextBias: z.array(z.string()).optional(),
},
  • src/tools-audio.ts:96-195 (registration)
    Registration of the voxtral_transcribe tool on the MCP server via server.registerTool, including title, description, input/output schemas, and annotations.
    server.registerTool(
  "voxtral_transcribe",
  {
    title: "Voxtral speech-to-text",
    description: [
      "Transcribe an audio file to text using Mistral Voxtral.",
      "",
      "Accepted models:",
      STT_MODELS.map((m) => `  - ${m}`).join("\n"),
      "",
      "Audio source is one of:",
      '  - { type: "file_url", fileUrl: "https://..." }  (public URL)',
      '  - { type: "file", fileId: "<id-from-files-api>" }',
      "",
      "Options:",
      "  - `language`: ISO-639-1 hint (e.g. 'fr', 'en'). Boosts accuracy when known.",
      "  - `temperature`: sampling temperature.",
      "  - `diarize`: return per-speaker segments (default false).",
      "  - `timestampGranularities`: ['segment'] to return per-segment timestamps.",
      "  - `contextBias`: list of phrases/terms that should bias the decoder.",
      "",
      "Returns plain `text`, detected `language`, optional `segments[]`, and token usage.",
    ].join("\n"),
    inputSchema: {
      audio: AudioSourceSchema,
      model: SttModelSchema.optional().describe(
        `STT model. Default: ${DEFAULT_STT_MODEL}.`
      ),
      language: z
        .string()
        .optional()
        .describe("ISO-639-1 language hint (e.g. 'fr', 'en')."),
      temperature: z.number().min(0).max(2).optional(),
      diarize: z.boolean().optional(),
      timestampGranularities: z
        .array(z.enum(["segment"]))
        .optional()
        .describe("Only 'segment' is currently supported."),
      contextBias: z.array(z.string()).optional(),
    },
    outputSchema: TranscribeOutputShape,
    annotations: {
      title: "Voxtral speech-to-text",
      readOnlyHint: true,
      destructiveHint: false,
      idempotentHint: true,
      openWorldHint: true,
    },
  },
  async (input) => {
    try {
      const model = input.model ?? DEFAULT_STT_MODEL;
      const request: {
        model: string;
        fileUrl?: string;
        fileId?: string;
        language?: string;
        temperature?: number;
        diarize?: boolean;
        timestampGranularities?: Array<"segment">;
        contextBias?: string[];
      } = { model };
      if (input.audio.type === "file_url") request.fileUrl = input.audio.fileUrl;
      if (input.audio.type === "file") request.fileId = input.audio.fileId;
      if (input.language !== undefined) request.language = input.language;
      if (input.temperature !== undefined) request.temperature = input.temperature;
      if (input.diarize !== undefined) request.diarize = input.diarize;
      if (input.timestampGranularities !== undefined)
        request.timestampGranularities = input.timestampGranularities;
      if (input.contextBias !== undefined) request.contextBias = input.contextBias;

      const res = await mistral.audio.transcriptions.complete(
        request as never
      );

      const segments = res.segments?.map((s) => ({
        text: s.text,
        start: s.start,
        end: s.end,
        score: s.score ?? undefined,
        speaker_id: s.speakerId ?? undefined,
      }));

      const structured = {
        text: res.text,
        model: res.model,
        language: res.language,
        segments,
        usage: mapUsage(res.usage),
      };

      return {
        content: [toTextBlock(res.text)],
        structuredContent: structured,
      };
    } catch (err) {
      return errorResult("voxtral_transcribe", err);
    }
  }
);

Tool Definition Quality

A4.3/5.0
Behavior3/5

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

Annotations already provide readOnlyHint, destructiveHint, idempotentHint, and openWorldHint. The description adds return details (text, language, segments, token usage) but does not disclose additional behavioral traits beyond what annotations cover.

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?

The description is well-structured with bullet points for models, audio sources, options, and returns. It is concise, front-loaded, and every sentence adds value.

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?

The description covers purpose, all parameters, and return data. It lacks mention of file size limits or supported formats beyond the schema, but overall it is thorough for a transcription tool.

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

Parameters5/5

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

With only 43% schema description coverage, the description compensates by explaining audio source structure, language hint, diarize, timestampGranularities, and contextBias. This adds significant meaning beyond the schema.

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

Purpose5/5

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

The description clearly states 'Transcribe an audio file to text using Mistral Voxtral,' providing a specific verb and resource. It distinguishes from sibling tools (e.g., codestral_fim, mistral_chat) which serve different purposes.

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?

The description provides clear usage context: accepted models, two audio source options, and explanation of optional parameters. However, it lacks explicit guidance on when not to use this tool or alternatives.

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

Install Server

Other Tools

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/Swih/mistral-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server