import { type AudioQuery, type SpeakResult, VoicevoxClient } from '@kajidog/voicevox-client'
import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
import type { CallToolResult } from '@modelcontextprotocol/sdk/types.js'
import * as z from 'zod/v4'
import { getConfig } from './config'
// 設定を取得
const config = getConfig()
// サーバー初期化
export const server = new McpServer({
name: 'MCP TTS Voicevox',
version: '0.3.1',
description: 'A Voicevox server that converts text to speech for playback and saving.',
})
// Voicevoxクライアント初期化
const voicevoxClient = new VoicevoxClient({
url: config.voicevoxUrl,
defaultSpeaker: config.defaultSpeaker,
defaultSpeedScale: config.defaultSpeedScale,
useStreaming: config.useStreaming,
})
// 無効化ツールのセット
const disabledTools = new Set(config.disabledTools)
// 制限設定
const restrictions = {
immediate: config.restrictImmediate,
waitForStart: config.restrictWaitForStart,
waitForEnd: config.restrictWaitForEnd,
}
// ユーティリティ関数
const createErrorResponse = (error: unknown): CallToolResult => ({
content: [
{
type: 'text',
text: `エラー: ${error instanceof Error ? error.message : String(error)}`,
},
],
})
const createSuccessResponse = (text: string): CallToolResult => ({
content: [{ type: 'text', text }],
})
const formatSpeakResponse = (result: SpeakResult): string => {
if (result.status === 'error') {
return `Error: ${result.errorMessage}`
}
const statusLabel = result.status === 'played' ? 'Played' : 'Queued'
const moreSegments = result.segmentCount > 1 ? ` +${result.segmentCount - 1} more` : ''
return `${statusLabel} (${result.mode}): "${result.textPreview}"${moreSegments}`
}
const parseAudioQuery = (query: string, speedScale?: number): AudioQuery => {
const audioQuery = JSON.parse(query) as AudioQuery
if (speedScale !== undefined) {
audioQuery.speedScale = speedScale
}
return audioQuery
}
const parseStringInput = (input: string): Array<{ text: string; speaker?: number }> => {
// \n と \\n の両方に対応するため、まず \\n を \n に変換してから分割
const normalizedInput = input.replace(/\\n/g, '\n')
const lines = normalizedInput.split('\n').filter((line) => line.trim())
return lines.map((line) => {
const match = line.match(/^(\d+):(.*)$/)
if (match) {
return { text: match[2].trim(), speaker: Number.parseInt(match[1], 10) }
}
return { text: line }
})
}
const processTextInput = async (
text: string,
speaker?: number,
speedScale?: number,
playbackOptions?: {
immediate?: boolean
waitForStart?: boolean
waitForEnd?: boolean
}
) => {
const segments = parseStringInput(text)
return await voicevoxClient.speak(segments, {
speaker,
speedScale,
...playbackOptions,
})
}
/**
* 条件付きツール登録（無効化されたツールは登録しない）
*/
function registerToolIfEnabled(name: string, definition: any, handler: any) {
if (disabledTools.has(name)) {
console.error(`Tool "${name}" is disabled via configuration`)
return
}
server.registerTool(name, definition, handler)
}
/**
* speak ツールの動的スキーマを構築
*/
function buildSpeakInputSchema() {
const schema: Record<string, z.ZodType> = {
text: z
.string()
.describe(
'Text split by line breaks (\\n). IMPORTANT: Each line = one speech unit (processed and played separately). Keep the FIRST LINE SHORT for quick playback start - audio begins as soon as the first line is synthesized. Example: "Hi!\\nThis is a longer explanation that follows." Optional speaker prefix per line: "1:Hello\\n2:World"'
),
query: z.string().optional().describe('Voice synthesis query'),
speaker: z.number().optional().describe('Default speaker ID (optional)'),
speedScale: z.number().optional().describe('Playback speed (optional, default from environment)'),
}
// 制限されていない場合のみスキーマに追加
if (!restrictions.immediate) {
schema.immediate = z
.boolean()
.optional()
.describe(
'If true, stops current playback and plays new audio immediately. If false, waits for current playback to finish. Default depends on environment variable.'
)
}
if (!restrictions.waitForStart) {
schema.waitForStart = z.boolean().optional().describe('Wait for playback to start (optional, default: false)')
}
if (!restrictions.waitForEnd) {
schema.waitForEnd = z.boolean().optional().describe('Wait for playback to end (optional, default: false)')
}
return schema
}
// ツール登録
// ping_voicevox ツール
registerToolIfEnabled(
'ping_voicevox',
{
title: 'Ping VOICEVOX',
description: 'Check if VOICEVOX Engine is running and reachable',
},
async (): Promise<CallToolResult> => {
try {
const health = await voicevoxClient.checkHealth()
if (health.connected) {
return createSuccessResponse(`VOICEVOX is running at ${health.url} (v${health.version})`)
}
return createErrorResponse(
new Error(`VOICEVOX is not reachable at ${health.url}. Please ensure VOICEVOX Engine is running.`)
)
} catch (error) {
return createErrorResponse(error)
}
}
)
// speak ツール
registerToolIfEnabled(
'speak',
{
title: 'Speak',
description:
'Convert text to speech and play it. Text is split by line breaks (\\n) into separate speech units. Each line is processed as an independent audio segment.',
inputSchema: buildSpeakInputSchema(),
},
async ({
text,
speaker,
query,
speedScale,
immediate,
waitForStart,
waitForEnd,
}: {
text: string
speaker?: number
query?: string
speedScale?: number
immediate?: boolean
waitForStart?: boolean
waitForEnd?: boolean
}): Promise<CallToolResult> => {
try {
// 設定からデフォルトの再生オプションを取得
const playbackOptions = {
immediate: immediate ?? config.defaultImmediate,
waitForStart: waitForStart ?? config.defaultWaitForStart,
waitForEnd: waitForEnd ?? config.defaultWaitForEnd,
}
let result: SpeakResult
if (query) {
const audioQuery = parseAudioQuery(query, speedScale)
result = await voicevoxClient.enqueueAudioGeneration(audioQuery, {
speaker,
speedScale,
...playbackOptions,
})
} else {
result = await processTextInput(text, speaker, speedScale, playbackOptions)
}
return createSuccessResponse(formatSpeakResponse(result))
} catch (error) {
return createErrorResponse(error)
}
}
)
registerToolIfEnabled(
'generate_query',
{
title: 'Generate Query',
description: 'Generate a query for voice synthesis',
inputSchema: {
text: z.string().describe('Text for voice synthesis'),
speaker: z.number().optional().describe('Default speaker ID (optional)'),
speedScale: z.number().optional().describe('Playback speed (optional, default from environment)'),
},
},
async ({
text,
speaker,
speedScale,
}: {
text: string
speaker?: number
speedScale?: number
}): Promise<CallToolResult> => {
try {
const query = await voicevoxClient.generateQuery(text, speaker, speedScale)
return createSuccessResponse(JSON.stringify(query))
} catch (error) {
return createErrorResponse(error)
}
}
)
registerToolIfEnabled(
'synthesize_file',
{
title: 'Synthesize File',
description: 'Generate an audio file and return its absolute path',
inputSchema: {
text: z
.string()
.optional()
.describe('Text for voice synthesis (if both query and text provided, query takes precedence)'),
query: z.string().optional().describe('Voice synthesis query'),
output: z.string().describe('Output path for the audio file'),
speaker: z.number().optional().describe('Default speaker ID (optional)'),
speedScale: z.number().optional().describe('Playback speed (optional, default from environment)'),
},
},
async ({
text,
query,
speaker,
output,
speedScale,
}: {
text?: string
query?: string
speaker?: number
output: string
speedScale?: number
}): Promise<CallToolResult> => {
try {
if (query) {
const audioQuery = parseAudioQuery(query, speedScale)
const filePath = await voicevoxClient.generateAudioFile(audioQuery, output, speaker)
return createSuccessResponse(filePath)
}
if (text) {
const filePath = await voicevoxClient.generateAudioFile(text, output, speaker, speedScale)
return createSuccessResponse(filePath)
}
throw new Error('queryパラメータとtextパラメータのどちらかを指定してください')
} catch (error) {
return createErrorResponse(error)
}
}
)
registerToolIfEnabled(
'stop_speaker',
{
title: 'Stop Speaker',
description: 'Stop current audio playback',
inputSchema: {
random_string: z.string().describe('Dummy parameter for no-parameter tools'),
},
},
async (): Promise<CallToolResult> => {
try {
await voicevoxClient.clearQueue()
return createSuccessResponse('スピーカーを停止しました')
} catch (error) {
return createErrorResponse(error)
}
}
)
registerToolIfEnabled(
'get_speakers',
{
title: 'Get Speakers',
description: 'Get a list of available speakers',
},
async (): Promise<CallToolResult> => {
try {
const speakers = await voicevoxClient.getSpeakers()
const result = speakers.flatMap((speaker: any) =>
speaker.styles.map((style: any) => ({
uuid: speaker.speaker_uuid,
speaker: style.id,
name: `${speaker.name}:${style.name}`,
}))
)
return createSuccessResponse(JSON.stringify(result))
} catch (error) {
return createErrorResponse(error)
}
}
)
registerToolIfEnabled(
'get_speaker_detail',
{
title: 'Get Speaker Detail',
description: 'Get detail of a speaker by id',
inputSchema: {
uuid: z.string().describe('Speaker UUID (speaker uuid)'),
},
},
async ({ uuid }: { uuid: string }): Promise<CallToolResult> => {
try {
const allSpeakers = await voicevoxClient.getSpeakers()
const targetSpeaker = allSpeakers.find((speaker: any) => speaker.speaker_uuid === uuid)
if (!targetSpeaker) {
throw new Error(`指定されたUUID ${uuid} のスピーカーが見つかりませんでした`)
}
const styles = targetSpeaker.styles.map((style: any) => ({
id: style.id,
name: style.name,
type: style.type || 'normal',
}))
const simplifiedInfo = {
uuid: targetSpeaker.speaker_uuid,
name: targetSpeaker.name,
version: targetSpeaker.version,
supported_features: targetSpeaker.supported_features,
styles: styles,
}
return createSuccessResponse(JSON.stringify(simplifiedInfo))
} catch (error) {
return createErrorResponse(error)
}
}
)
// 設定エクスポート（テスト用）
export { config }
