Transcribe Audio URL

Instructions

Implementation Reference

• src/tools/transcribe-url.ts:31-37 (handler)

  The handler function that executes the transcribe_audio_url tool logic. Calls client.submitTranscription(url, language) and returns the job_id.

  async ({ url, language }) => {
    const r = await client.submitTranscription(url, language);
    return {
      content: [{ type: 'text', text: `Job submitted. job_id: ${r.id}` }],
      structuredContent: jsonStructured(r),
    };
  },

• src/tools/transcribe-url.ts:10-29 (schema)

  Input schema for transcribe_audio_url: requires a public audio/video URL (max 200 MB) and optional 2-letter ISO language code.

  {
    title: 'Transcribe Audio URL',
    description:
      'Submit a publicly-accessible audio or video URL for transcription with speaker diarization. Returns a job_id you can poll with get_transcription_status, or use transcribe_audio_url_sync to wait synchronously.',
    inputSchema: z.object({
      url: z
        .string()
        .url()
        .describe(
          'Public URL of audio/video file (mp3, m4a, wav, mp4, webm, etc.) Max 200 MB.',
        ),
      language: z
        .string()
        .length(2)
        .optional()
        .describe(
          'ISO 639-1 code, e.g. "en", "es". Auto-detected if omitted.',
        ),
    }),
    annotations: { readOnlyHint: false, openWorldHint: true },

• src/tools/transcribe-url.ts:7-39 (registration)

  Registration function that calls server.registerTool with name 'transcribe_audio_url', schema, annotations, and handler.

  export function register(server: McpServer, client: GhostMinutesClient): void {
    server.registerTool(
      'transcribe_audio_url',
      {
        title: 'Transcribe Audio URL',
        description:
          'Submit a publicly-accessible audio or video URL for transcription with speaker diarization. Returns a job_id you can poll with get_transcription_status, or use transcribe_audio_url_sync to wait synchronously.',
        inputSchema: z.object({
          url: z
            .string()
            .url()
            .describe(
              'Public URL of audio/video file (mp3, m4a, wav, mp4, webm, etc.) Max 200 MB.',
            ),
          language: z
            .string()
            .length(2)
            .optional()
            .describe(
              'ISO 639-1 code, e.g. "en", "es". Auto-detected if omitted.',
            ),
        }),
        annotations: { readOnlyHint: false, openWorldHint: true },
      },
      async ({ url, language }) => {
        const r = await client.submitTranscription(url, language);
        return {
          content: [{ type: 'text', text: `Job submitted. job_id: ${r.id}` }],
          structuredContent: jsonStructured(r),
        };
      },
    );
  }

• src/server.ts:32-42 (registration)

  Top-level registration: calls registerTranscribeUrl(server, client) to wire up the tool on the server.

    registerTranscribeUrl(server, client);
    registerTranscribeSync(server, client);
    registerGetStatus(server, client);
    registerListTranscripts(server, client);
    registerGetTranscript(server, client);
    registerDeleteTranscript(server, client);
    registerSummarize(server, client);
    registerGetCredits(server, client);
  
    return server;
  }

• src/client.ts:68-92 (helper)

  The GhostMinutesClient.submitTranscription method which POSTs to /mcp/transcribe with audio_file_url and optional language.

  async submitTranscription(
    audioUrl: string,
    language?: string,
  ): Promise<{ id: string; [key: string]: unknown }> {
    try {
      const res = await this.http.post<
        unknown,
        AxiosResponse<{ id: string; [key: string]: unknown }>
      >(
        '/mcp/transcribe',
        {
          audio_file_url: audioUrl,
          ...(language ? { language } : {}),
        },
        {
          headers: this.hasApiKey()
            ? { Authorization: `Bearer ${this.apiKey}` }
            : {},
        },
      );
      return this.ensureOk(res);
    } catch (e) {
      this.handleThrown(e);
    }
  }