@vapi-ai/mcp-server
Version:
Vapi MCP Server
295 lines (294 loc) • 8.73 kB
JavaScript
import { z } from 'zod';
// ===== Model Provider and Models =====
export const ModelProvider = {
OpenAI: 'openai',
Anthropic: 'anthropic',
GoogleAI: 'google',
};
export const OpenAIModels = {
GPT4o: 'gpt-4o',
GPT4oMini: 'gpt-4o-mini',
};
export const AnthropicModels = {
Claude3Sonnet: 'claude-3-7-sonnet-20250219',
Claude3Haiku: 'claude-3-5-haiku-20241022',
};
export const GoogleModels = {
GeminiPro: 'gemini-1.5-pro',
GeminiFlash: 'gemini-1.5-flash',
Gemini2Flash: 'gemini-2.0-flash',
Gemini2Pro: 'gemini-2.0-pro',
};
const OpenAILLMSchema = z.object({
provider: z.literal(ModelProvider.OpenAI),
model: z.enum([OpenAIModels.GPT4o, OpenAIModels.GPT4oMini]),
});
const AnthropicLLMSchema = z.object({
provider: z.literal(ModelProvider.Anthropic),
model: z.enum([
AnthropicModels.Claude3Sonnet,
AnthropicModels.Claude3Haiku,
]),
});
const GoogleLLMSchema = z.object({
provider: z.literal(ModelProvider.GoogleAI),
model: z.enum([
GoogleModels.GeminiPro,
GoogleModels.GeminiFlash,
GoogleModels.Gemini2Flash,
GoogleModels.Gemini2Pro,
]),
});
const GenericLLMSchema = z.object({
provider: z.string(),
model: z.string(),
});
const LLMSchema = z.union([
OpenAILLMSchema,
AnthropicLLMSchema,
GoogleLLMSchema,
GenericLLMSchema,
]);
export const DEFAULT_LLM = {
provider: ModelProvider.OpenAI,
model: OpenAIModels.GPT4o,
};
const VoiceProviderSchema = z.enum([
'vapi',
'11labs',
'azure',
'cartesia',
'custom-voice',
'deepgram',
'11labs',
'hume',
'lmnt',
'neuphonic',
'openai',
'playht',
'rime-ai',
'smallest-ai',
'tavus',
'sesame',
]);
export const DEFAULT_VOICE = {
provider: '11labs',
voiceId: 'sarah',
};
export const DEFAULT_TRANSCRIBER = {
provider: 'deepgram',
model: 'nova-3',
};
export const ElevenLabsVoiceIds = {
Sarah: 'sarah',
Phillip: 'phillip',
Steve: 'steve',
Joseph: 'joseph',
Myra: 'myra',
};
// ===== Common Schemas =====
export const BaseResponseSchema = z.object({
id: z.string(),
createdAt: z.string(),
updatedAt: z.string(),
});
// ===== Assistant Schemas =====
export const CreateAssistantInputSchema = z.object({
name: z.string().describe('Name of the assistant'),
instructions: z
.string()
.optional()
.default('You are a helpful assistant.')
.describe('Instructions for the assistant'),
llm: z
.union([
LLMSchema,
z.string().transform((str) => {
try {
return JSON.parse(str);
}
catch (e) {
throw new Error(`Invalid LLM JSON string: ${str}`);
}
}),
])
.default(DEFAULT_LLM)
.describe('LLM configuration'),
toolIds: z
.array(z.string())
.optional()
.describe('IDs of tools to use with this assistant'),
transcriber: z
.object({
provider: z.string().describe('Provider to use for transcription'),
model: z.string().describe('Transcription model to use'),
})
.default(DEFAULT_TRANSCRIBER)
.describe('Transcription configuration'),
voice: z
.object({
provider: VoiceProviderSchema.describe('Provider to use for voice'),
voiceId: z.string().describe('Voice ID to use'),
model: z.string().optional().describe('Voice model to use'),
})
.default(DEFAULT_VOICE)
.describe('Voice configuration'),
firstMessage: z
.string()
.optional()
.default('Hello, how can I help you today?')
.describe('First message to say to the user'),
firstMessageMode: z
.enum([
'assistant-speaks-first',
'assistant-waits-for-user',
'assistant-speaks-first-with-model-generated-message',
])
.default('assistant-speaks-first')
.optional()
.describe('This determines who speaks first, either assistant or user'),
});
export const AssistantOutputSchema = BaseResponseSchema.extend({
name: z.string(),
llm: z.object({
provider: z.string(),
model: z.string(),
}),
voice: z.object({
provider: z.string(),
voiceId: z.string(),
model: z.string().optional(),
}),
transcriber: z.object({
provider: z.string(),
model: z.string(),
}),
toolIds: z.array(z.string()).optional(),
});
export const GetAssistantInputSchema = z.object({
assistantId: z.string().describe('ID of the assistant to get'),
});
export const UpdateAssistantInputSchema = z.object({
assistantId: z.string().describe('ID of the assistant to update'),
name: z.string().optional().describe('New name for the assistant'),
instructions: z
.string()
.optional()
.describe('New instructions for the assistant'),
llm: z
.union([
LLMSchema,
z.string().transform((str) => {
try {
return JSON.parse(str);
}
catch (e) {
throw new Error(`Invalid LLM JSON string: ${str}`);
}
}),
])
.optional()
.describe('New LLM configuration'),
toolIds: z
.array(z.string())
.optional()
.describe('New IDs of tools to use with this assistant'),
transcriber: z
.object({
provider: z.string().describe('Provider to use for transcription'),
model: z.string().describe('Transcription model to use'),
})
.optional()
.describe('New transcription configuration'),
voice: z
.object({
provider: VoiceProviderSchema.describe('Provider to use for voice'),
voiceId: z.string().describe('Voice ID to use'),
model: z.string().optional().describe('Voice model to use'),
})
.optional()
.describe('New voice configuration'),
firstMessage: z
.string()
.optional()
.describe('First message to say to the user'),
firstMessageMode: z
.enum([
'assistant-speaks-first',
'assistant-waits-for-user',
'assistant-speaks-first-with-model-generated-message',
])
.optional()
.describe('This determines who speaks first, either assistant or user'),
});
// ===== Call Schemas =====
export const CallInputSchema = z.object({
assistantId: z
.string()
.optional()
.describe('ID of the assistant to use for the call'),
phoneNumberId: z
.string()
.optional()
.describe('ID of the phone number to use for the call'),
customer: z
.object({
number: z.string().describe('Customer phone number'),
})
.optional()
.describe('Customer information'),
scheduledAt: z
.string()
.optional()
.describe('ISO datetime string for when the call should be scheduled (e.g. "2025-03-25T22:39:27.771Z")'),
assistantOverrides: z
.object({
variableValues: z
.record(z.string(), z.any())
.optional()
.describe('Key-value pairs for dynamic variables to use in the assistant\'s prompts (e.g. {"name": "Joe", "age": "24"})'),
})
.optional()
.describe('Overrides for the assistant configuration'),
});
export const CallOutputSchema = BaseResponseSchema.extend({
status: z.string(),
endedReason: z.string().optional(),
assistantId: z.string().optional(),
phoneNumberId: z.string().optional(),
customer: z
.object({
number: z.string(),
})
.optional(),
scheduledAt: z.string().optional(),
});
export const GetCallInputSchema = z.object({
callId: z.string().describe('ID of the call to get'),
});
// ===== Phone Number Schemas =====
export const GetPhoneNumberInputSchema = z.object({
phoneNumberId: z.string().describe('ID of the phone number to get'),
});
export const PhoneNumberOutputSchema = BaseResponseSchema.extend({
name: z.string().optional(),
phoneNumber: z.string(),
status: z.string(),
capabilities: z
.object({
sms: z.boolean().optional(),
voice: z.boolean().optional(),
})
.optional(),
});
// ===== Tool Schemas =====
export const GetToolInputSchema = z.object({
toolId: z.string().describe('ID of the tool to get'),
});
export const ToolOutputSchema = BaseResponseSchema.extend({
type: z
.string()
.describe('Type of the tool (dtmf, function, mcp, query, etc.)'),
name: z.string().describe('Name of the tool'),
description: z.string().describe('Description of the tool'),
});