April 2025 Summaries

In April, the focus was on enhancing Vapi's platform by introducing tools to improve integration, support early-stage voice AI startups, and refine core functionalities based on user feedback. The new features include Vapi's support for the Model Context Protocol (MCP), enabling seamless integration with applications like Zapier and Composio, and allowing Vapi agents to send and receive SMS messages via Twilio. A Vapi Startups Program was launched to aid early-stage voice AI entrepreneurs with resources like free minutes, priority feature access, and a Slack community. The platform now supports new language models and voices, including o3-mini, o4-mini, and enhanced transcription with Deepgram Nova-3. Platform enhancements include pre-answer call hooks, batch calling, new built-in tools for Slack and Google integrations, multi-structured data extraction, and WebSocket call support, aiming to streamline the development and debugging of large-scale voice systems. Users are encouraged to share their challenges and future needs to guide ongoing improvements.

Plivo's integration with Vapi via SIP offers a streamlined approach for scaling voice capabilities without losing flexibility, especially for those using Plivo for telephony who wish to enhance their voice stack with real-time language models. This integration allows users to treat SIP as a programmable transport layer, facilitating the use of existing Plivo numbers with Vapi to power multilingual voice agents built on modern language models. It supports both inbound and outbound calls, granting developers full control over routing and model selection while maintaining existing infrastructure. Setting up outbound calls involves creating SIP trunks in Plivo, whitelisting Vapi's IPs, and using Vapi's API for call management, while inbound calls require configuring IP URIs and SIP endpoints. This setup is designed to ensure that telephony does not hinder language model workflows, allowing for open, programmable, and scalable voice operations, and it is compatible with any SIP-compatible carrier, although it is particularly beneficial for current Plivo users.

Vapi has introduced SMS functionality, allowing AI voice agents to send and receive text messages, thus enhancing their multimodal capabilities and aligning more closely with real-world communication needs. With this new feature, agents can seamlessly switch from voice to text during conversations, support text-based interactions, and send outbound messages or respond to incoming SMS, which is currently available through Twilio integration. This addition is designed to improve versatility and reliability in environments such as healthcare, delivery, and retail by providing alternative communication methods for confirming appointments, collecting input, and recovering from call disruptions. The rollout aims for greater coverage and control, with future plans to support more providers and enhance SMS tools with features like MMS, message templates, and integrated voice-text conversation histories.

The Vapi Model Context Protocol (MCP) Server is a new tool that allows users to control voice agents through function calling from any AI agent tool, enabling real-time voice flows, outbound calls, and routing to Vapi assistants without leaving the chat interface. This integration, which supports both local and remote setups, enhances workflow efficiency by merging large language models (LLMs) with automation, allowing users to execute tasks like scheduling follow-ups or making outbound calls using natural language. The MCP server is already being utilized in various sectors, such as healthcare and sales, to streamline communication processes by allowing agents like Claude Desktop to trigger voice assistants and perform tasks with minimal context switching. The server can be configured on Claude Desktop or accessed remotely, making it easy to initiate voice workflows and providing a foundational step towards more advanced multi-turn assistant control and seamless agent orchestration between chat and voice interactions.

Aura-2, Deepgram's latest text-to-speech model, is now available on Vapi, offering improved voice quality, pronunciation accuracy, and low latency for applications such as outbound sales, AI-powered IVRs, and healthcare assistants. Unlike traditional TTS models, Aura-2 is trained on human conversational data, allowing for more natural and context-aware speech that mirrors human interactions, complete with appropriate pauses and emotional tone. It has been tested across various industries like healthcare and finance to ensure precision and clarity, especially in complex terminology. Aura-2 supports real-time, low-latency performance suitable for dynamic interactions, making it ideal for use cases like customer support, sales, and interactive voice menus. Switching to Aura-2 on Vapi is straightforward, requiring no extra integration, and it optimally utilizes Deepgram's latest API version for seamless conversational experiences.

Vapi MCP client introduces a transformative approach to building voice agents that seamlessly integrate with Model Context Protocol (MCP) servers, allowing immediate connectivity with external systems like Zapier or Composio without the need for individual API integrations. By supporting JSON-RPC over HTTP/2, tool schema validation, and automatic retries, Vapi MCP reduces development complexity and enhances the capabilities of voice agents by dynamically adapting to available tools and context during interactions. This integration facilitates tasks such as scheduling, CRM updates, and real-time data fetching while enabling agents to access thousands of pre-built actions across various applications. Users can easily set up Vapi by obtaining an MCP server URL, creating a tool in the Vapi Dashboard, and linking it to their assistant, which can then leverage the tools provided by the MCP server to perform actions like checking order statuses or rescheduling appointments with simple voice commands.

Christian Ulstrup, founder of GSD at Work, revolutionized his consulting practice by integrating Vapi, a voice-driven AI tool, to streamline the intake process for his workshops. Faced with the challenge of gathering deep, contextual insights from global clients without the logistical hurdles of scheduling live calls, Ulstrup adopted Vapi to enable participants to asynchronously provide voice-based responses. This system captures transcripts and summaries, which are then analyzed to customize workshop agendas, resulting in highly personalized and efficient sessions. The implementation of Vapi has not only enhanced the quality and impact of his AI Action Sessions but also earned high client satisfaction, demonstrating the effectiveness of blending efficiency with personalization. Ulstrup continues to explore new applications for this voice-powered model, positioning it as a critical element of his consulting toolkit.