Native MCP Server · OAuth 2.0

Give your AI assistant a phone line

Connect Voiceyfy to Claude, ChatGPT, Gemini, Llama, DeepSeek or Mistral. Ask in plain English and your assistant places the call, reads the transcript, books the appointment and sends the WhatsApp, all inside your own account, with the permissions you approve.

Claude

ChatGPT

Google Gemini

Meta Llama

DeepSeek

Mistral

Disclaimer: Voiceyfy is an independent AI platform. All product names, trademarks, and logos (ChatGPT, Claude, Gemini, DeepSeek, Llama, Mistral) belong to their respective owners. Voiceyfy is not affiliated with or endorsed by OpenAI, Anthropic, Google, Meta, DeepSeek, or Mistral AI.

Start Free Trial See how to connect

Your AI assistant

Call the three leads who enquired yesterday and book anyone who is interested.

list_leads3 leads found

schedule_callcalling +91 93476…

list_appointments2 slots free Thu

Called all three. Priya booked Thursday 4pm, Arjun asked for a callback Monday, and the third number did not answer, so I have queued a retry.

AI platforms supported

Tools exposed to your model

Transports: MCP & OpenAPI

OAuth 2.0

Scoped, revocable access

Every major model, one connection

Whichever assistant your team already lives in, Voiceyfy plugs into it. MCP-capable clients discover the tools automatically; the rest import the same capabilities from an OpenAPI spec.

Claude

Custom MCP Connector

Add Voiceyfy as a custom connector in Claude.ai or Claude Desktop and ask it to call, book and follow up in plain English.

Connects via MCP server

ChatGPT

Custom GPT Action

Build a Custom GPT that dials your customers, reads transcripts and books appointments from your Voiceyfy account.

Connects via OpenAPI spec

Google Gemini

Function Calling / Custom Tool

Wire Voiceyfy into Google AI Studio or Vertex AI as a function-calling tool for your Gemini agents.

Connects via MCP server

Meta Llama

MCP / REST Tool

Run Llama locally in Ollama, AnythingLLM or OpenWebUI and give it the same Voiceyfy tools the hosted models get.

Connects via MCP server

DeepSeek

MCP HTTP Transport

Give DeepSeek R1 and V3 reasoning agents a phone line, a calendar and a CRM through the MCP HTTP transport.

Connects via MCP server

Mistral

Agent Action

Import the OpenAPI spec as a custom tool in Le Chat or La Plateforme and let Mistral agents dispatch real calls.

Connects via OpenAPI spec

What your model can actually do

Thirteen tools, published the moment you connect. These are the real names your assistant sees in its tool list. No wrapper to write, no schema to maintain.

Agents

Find the right voice agent and read its configuration before doing anything else.

list_agents
List every voice agent on the account with its id, name and call mode. This is what turns an agent name into the id the other tools need.
get_agent
Read one agent in full: its greeting, prompt, voice, language and the tools it is allowed to use on a call.

Calls

Search call history, read complete transcripts, and place real outbound calls.

list_calls
Browse recent calls with their outcome, duration and AI summary, filtered by agent, date or phone number.
get_call
Everything about a single call, including the complete verbatim transcript of what both sides actually said.
schedule_callTakes real action
Places a real phone call. An AI agent dials the recipient and speaks to them, now or at a scheduled time.

Leads & CRM

Ask who a caller is, whether they are already in the CRM, and what happened with them.

list_leads
List CRM leads and contacts newest first, with name, phone, stage and source, and check whether a number is already known.
get_lead
One lead in full, including the CRM fields the list view leaves out: notes, custom fields and activity.

Insights & Numbers

Aggregate answers to how many, how long and which agent, plus what is on the calendar.

get_call_analytics
Aggregate call statistics over a recent period: volume, talk time, outcomes and per-agent breakdowns.
list_appointments
Appointments booked through the account, most recent first. Use it for what is booked, who is coming and no-show checks.
list_phone_numbers
The phone numbers connected to the account, what each one is used for and which agent answers it.

Draft templates, get them through Meta review, and send approved messages.

list_whatsapp_templates
List WhatsApp message templates with the status Meta has given each one. Only APPROVED templates can be sent.
create_whatsapp_templateTakes real action
Create a template and submit it to Meta for review. This registers wording for approval; it does not message anyone.
send_whatsapp_templateTakes real action
Sends a real WhatsApp message to a person, using an approved template.

Ask in English. It runs in your account.

You never name a tool. The model picks the right one, chains them when it needs to, and comes back with the answer.

“How did our calls go this week?”

get_call_analytics returns volume, talk time and per-agent outcomes for the period.

“What exactly did the customer say on yesterday's 4pm call?”

get_call returns the complete verbatim transcript, not a summary of it.

“Call Priya back about her AC service and confirm Thursday.”

list_leads finds Priya, then schedule_call dials her with your service agent.

“Who is booked in on Thursday, and did anyone no-show last week?”

list_appointments returns the bookings with their status and history.

“Send the approved appointment reminder to today's bookings.”

list_whatsapp_templates checks approval, send_whatsapp_template delivers it.

“Which number does our sales agent answer on?”

list_phone_numbers maps every connected number to the agent behind it.

Two ways in, one set of capabilities

Both endpoints sit behind the same OAuth 2.0 authorisation server and the same scopes. Pick whichever one your client understands.

MCP Server

Streamable HTTP · OAuth 2.0

The native path. Your client discovers every Voiceyfy tool automatically, so new tools appear without you touching the configuration again.

MCP Server URL

https://secure.voiceyfy.com/mcp

Claude · Gemini · Llama · DeepSeek

OpenAPI Actions

OpenAPI 3 · OAuth 2.0

For clients that build tools from a REST specification. Import the URL once and the same operations arrive as callable actions.

OpenAPI Spec URL

https://secure.voiceyfy.com/openapi.yaml

ChatGPT · Mistral

Authorization URL

https://secure.voiceyfy.com/authorize

Token URL

https://secure.voiceyfy.com/api/oauth/token

Connect in about two minutes

Pick your platform. The steps below are the same four you will see in the dashboard under Settings → Developer.

Connect Claude

Custom MCP Connector · OAuth 2.0 · takes about two minutes

MCP Server URL

https://secure.voiceyfy.com/mcp

1
Open Claude Settings
Go to Settings → Connectors (or Developer → Custom Connectors) in Claude.ai or Claude Desktop.
2
Add a custom MCP connector
Click Add Custom Connector, or Add MCP Server, depending on your Claude version.
3
Paste your credentials
Paste the MCP Server URL, Client ID and Client Secret from Settings → Developer in your Voiceyfy dashboard.
4
Authorize the connection
Click Connect and approve access to your Voiceyfy agents and calendar workspace. Claude can now use your tools.

Your Client ID and Client Secret live in the Voiceyfy dashboard under Settings → Developer. Secrets are shown once, when the key is created.

Scoped access, not a master key

You decide what the model is allowed to touch

Every connection is its own OAuth 2.0 client with its own ID and secret. Before the first tool ever runs, you approve the exact scopes on a consent screen, so a research assistant can read call history without ever being able to dial a customer.

Separate credentials per client
Connect Claude and ChatGPT with different keys. Revoke one without touching the other.
Explicit consent before first use
The consent screen lists every scope being requested. Nothing runs until you approve it.
Every tool call is audited
Each call is logged against the connection that made it, with its arguments and outcome.
Write actions are held separately
Placing calls and sending WhatsApp messages are their own scopes, so read access can never do either.

Scopes you can grant

agents:readRead voice and chat agent configuration
call_activity:readRead call history and transcripts
scheduled_calls:readRead scheduled and queued calls
scheduled_calls:writePlace and schedule outbound calls
leads:readRead CRM leads and contacts
leads:writeCreate and update CRM leads
customers:readRead customer records
customers:writeCreate and update customer records
calendar:readRead appointments and availability
calendar:writeBook, reschedule and cancel appointments
knowledge_bases:readRead knowledge base content
knowledge_bases:writeAdd and update knowledge base content
chat_activity:readRead chat and WhatsApp conversations
phone_numbers:readRead connected phone numbers

And the models that run Voiceyfy itself

The integration works both ways. Point your instance at the provider you already trust, with your own API keys, and the whole platform follows.

OpenAI

The default engine. GPT-4o and GPT-4o-mini power prompt generation, call summaries, key-term extraction and knowledge base cleaning.

Google Gemini

Add your Gemini API key to route text generation through Google's models instead, without changing anything else in your setup.

Anthropic Claude

Add your Anthropic key to run the same text generation on Claude. The choice is per instance, so your data stays with the vendor you picked.

One setting, whole platform

The model you choose drives prompt generation, call summaries, key-term extraction, policy drafting and self-training alike.

Retrieval on your own content

Knowledge base answers come from OpenAI embeddings stored in a Pinecone vector index, so agents quote your material, not the model's general knowledge.

Speech stays purpose-built

Transcription and voice run on dedicated speech models tuned for Indian accents and 22 languages, independent of your text model choice.

LLM Integration FAQs

What is an LLM integration in Voiceyfy?

It is a secure connection between your Voiceyfy account and an AI assistant such as Claude, ChatGPT, Gemini, Llama, DeepSeek or Mistral. Once connected, that assistant can use Voiceyfy tools directly in the chat: list your agents, read call transcripts, place an outbound call, check what is booked and send an approved WhatsApp template. You ask in plain English; the model does the work in your account.

What is MCP, and why does Voiceyfy use it?

MCP, the Model Context Protocol, is the open standard AI clients use to discover and call external tools. Voiceyfy ships a native MCP server over streamable HTTP with OAuth 2.0, so any MCP-capable client (Claude, Ollama, AnythingLLM, OpenWebUI, and LangChain or LlamaIndex agents) can connect without a custom adapter. Clients that do not speak MCP, such as ChatGPT Custom GPTs and Mistral agents, import the same capabilities from an OpenAPI specification instead.

Which AI platforms can I connect today?

Six: Claude, ChatGPT, Google Gemini, Meta Llama, DeepSeek and Mistral. Claude, Gemini, Llama and DeepSeek connect through the MCP server URL. ChatGPT and Mistral import the OpenAPI spec and authenticate with the same OAuth 2.0 client credentials.

Is it safe to let an AI assistant into my Voiceyfy account?

Every connection is an OAuth 2.0 client with its own ID and secret, and you approve exactly which scopes it gets on a consent screen before the first tool runs. A research assistant can be given read-only scopes so it can never place a call or message a customer. Actions that reach a real person, such as placing a call or sending a WhatsApp message, are separate write scopes, every tool call is logged against the connection that made it, and you can revoke a client from Settings → Developer at any time.

Can my AI assistant actually place a phone call?

Yes, if you grant it the scope to. The schedule_call tool dials a real number with one of your Voiceyfy voice agents, immediately or at a time you specify, and the agent speaks to the person who answers. Because it takes a real-world action, the tool is deliberately explicit: models are instructed to use it only when a call has clearly been asked for, never on a hypothetical question.

Do I need to write any code?

No. Create an API key in Settings → Developer, copy the server URL, Client ID and Client Secret, and paste them into your AI client's connector screen. The whole setup is four steps and takes about two minutes. Developers who want to go further can call the same capabilities directly over the REST API and webhooks.

Which models run inside Voiceyfy itself?

Voiceyfy runs on GPT-class models by default, and your instance can be pointed at OpenAI, Google Gemini or Anthropic Claude with your own API keys. The same setting controls prompt generation, call summaries, knowledge base cleaning and self-training. Knowledge base retrieval uses OpenAI embeddings stored in a Pinecone vector index, so your agents answer from your own content rather than the model's general knowledge.

Keep exploring

Business tool integrations API & webhooks Knowledge base Trust & security

Put your AI assistant to work on the phones

Start a 14-day free trial, create an API key, and connect your first model in the time it takes to make a coffee.

Start Free Trial Talk to us