Indian artificial intelligence company Sarvam AI has launched Voice Agents MCP, a hosted integration that allows compatible AI assistants and coding tools to create, configure, test and manage voice agents through natural-language instructions. The new server connects Sarvam’s voice-agent platform with tools such as Claude, Cursor and Visual Studio Code (VS Code), allowing developers to work with voice automation without repeatedly switching between dashboards, API documentation and development environments.
The launch expands the use of the Model Context Protocol (MCP), a standard that allows AI applications to connect to external tools and services. Through Sarvam Voice Agents MCP, developers can manage voice-agent configurations, connect telephony, launch outbound calling campaigns, create knowledge bases, run evaluations and inspect call analytics. The move could simplify the development of automated customer support and business communication systems, particularly for companies building voice experiences in Indian languages.
Key Takeaways
- New launch: Sarvam AI has introduced Voice Agents MCP, a hosted server connecting AI assistants to its voice-agent platform.
- Compatible tools: Developers can connect supported clients such as Claude, Cursor and VS Code.
- Agent management: The integration supports agent configuration, testing, telephony, campaigns, knowledge bases and analytics.
- Natural-language workflow: Developers can describe tasks to their coding assistant instead of manually navigating every platform feature.
- Indian-language focus: Sarvam’s broader voice-agent platform supports 11 Indian languages plus English.
- Remote connection: The MCP server is hosted by Sarvam, so developers do not need to install and run the server locally.
- Business applications: Potential uses include appointment reminders, customer support, payment follow-ups and lead generation.
What Is Sarvam Voice Agents MCP?
Sarvam Voice Agents MCP is an integration between Sarvam’s voice-agent platform and AI applications that support the Model Context Protocol.
MCP provides a standard way for AI applications to discover and use external tools. Instead of relying only on information contained in a model’s training data, an AI assistant can use an authorised connection to interact with a service and carry out specific operations.
In this case, the external service is Sarvam’s voice-agent platform. Developers can ask a compatible assistant to perform tasks involving their agents, campaigns, knowledge bases and call reports. The assistant can then use the tools exposed by the MCP server to complete supported operations.
For example, a developer could describe a customer-support agent that answers common questions, uses a supplied knowledge base and follows a defined conversation flow. The coding assistant could help configure that agent through Sarvam’s platform rather than requiring the developer to locate every setting manually.
The integration does not eliminate the underlying voice infrastructure. Sarvam still provides the services used to build and operate agents, while the MCP server gives AI assistants a standard interface for working with those services.
The distinction is important: MCP is the connection layer, not a voice model or a replacement for the complete agent platform.
How the New MCP Server Works
Sarvam provides Voice Agents MCP as a hosted remote server at https://mcp.sarvam.ai/voice-agents. Developers register the endpoint in a compatible client and authenticate through Sarvam’s sign-in process.
Once connected, the assistant can discover the tools available to the authenticated workspace and use them to perform supported actions.
The workflow can be understood in four stages:
- Connect: Register Sarvam’s remote MCP server in a supported AI client.
- Authenticate: Sign in through Sarvam’s single sign-on process and approve the connection.
- Describe the task: Tell the assistant what kind of voice agent or operation is required.
- Review and test: Inspect the configuration, run evaluations and verify the agent’s behaviour before deployment.
Sarvam’s documentation describes the server as a hosted service, meaning developers do not need to install the server itself on their computers. However, they still need a Sarvam account with access to Voice Agents and a compatible MCP client.
The connection is associated with the organisation and workspace resolved by the user’s login. To work in another workspace, users need to reconnect with an account for that workspace.
This design makes the integration convenient for existing development workflows, while keeping access tied to the authenticated account and its available permissions.
What Developers Can Build and Manage
The integration exposes several areas of Sarvam’s voice-agent platform to compatible AI assistants.
| Capability | What it enables |
|---|---|
| Agent configuration | Create, clone and configure voice agents, including prompts, variables, tools and voice settings |
| Telephony | Connect providers, onboard phone numbers and deploy agents for incoming calls |
| Outbound campaigns | Prepare campaigns, upload contact groups and manage campaign operations |
| Knowledge bases | Create knowledge collections, upload files and retrieve relevant information |
| Evaluations | Define test scenarios and check agent behaviour against expected outcomes |
| Analytics | Review reports, interaction traces, call replays and performance insights |
| Analytics boards | Build dashboards to examine operational data and track selected metrics |
These capabilities cover much of the lifecycle involved in building a production voice agent. Developers can work on the agent’s behaviour, test it, deploy it and examine how it performs without treating each step as an entirely separate workflow.
The exact tools available can depend on the user’s workspace, account permissions and plan. Sarvam’s documentation notes that operations or fields unavailable to a particular workspace may not appear in the tool interface.
Why Sarvam Is Connecting Voice Agents to Coding Assistants
Building a reliable voice agent involves more than writing a prompt. Developers may need to configure speech recognition, select a voice, define conversation flows, connect business systems, set up telephony and establish a testing process.
When these activities are spread across multiple tools, development can become slow. A developer may need to consult documentation, switch between interfaces, copy settings and repeatedly test changes.
Connecting a coding assistant to the platform offers a different approach. The developer describes the intended outcome, while the assistant uses the available tools to perform the relevant configuration steps.
This is particularly useful for developers who already use AI-assisted coding environments. They can work on an application and its voice-agent functionality within a more consistent workflow.
However, natural-language instructions do not guarantee a correct result. Developers still need to verify configuration choices, test the agent with realistic conversations and confirm that integrations behave as expected. Production deployments may also require human approval, especially when they involve customer data, payments or outbound calls.
India’s Multilingual Voice AI Opportunity
Sarvam’s broader Voice Agents platform is designed for businesses operating in India. Its documentation lists support for 11 Indian languages plus English: Hindi, Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Odia and Assamese.
The platform is designed to handle code-mixed speech, names, numbers and other elements that can be difficult for conversational systems. These capabilities are important in Indian customer interactions, where callers may switch between English and a regional language within the same conversation.
The platform combines speech recognition, language-model reasoning and speech synthesis with telephony, testing and analytics. Sarvam says the models in its voice-agent stack are self-hosted, with data residency in India. That may be relevant to businesses with data-handling and compliance requirements, although individual organisations should still evaluate their obligations and the precise terms of service.
The MCP integration does not itself add language support. Instead, it gives AI coding assistants a way to configure and operate the voice-agent capabilities Sarvam already offers.
For businesses, this could make it easier to prototype multilingual agents for customer service, appointment booking, lead qualification and payment reminders. The commercial value will depend on how reliably those agents understand callers, handle exceptions and complete tasks.
Potential Business Uses
The ability to manage voice agents through an AI assistant could be useful across several business workflows.
Customer support
A business could configure an agent to answer frequently asked questions, retrieve information from a knowledge base and route more complex issues to human staff. Evaluation tools can help teams test common scenarios before deployment.
Appointment reminders
Healthcare providers, service businesses and other organisations could use voice agents to remind customers about appointments, confirm attendance and explain rescheduling options. Such deployments need appropriate consent, privacy controls and escalation paths.
Payment follow-ups
Businesses may use automated calls to remind customers about upcoming or overdue payments. These workflows require careful controls to avoid disclosing sensitive financial information to the wrong person or contacting customers inappropriately.
Lead generation and outreach
Sales teams can use outbound campaigns to contact prospective customers and collect responses. Campaign performance can then be examined through call analytics. Organisations must ensure that their outreach complies with applicable telemarketing, consent and privacy rules.
In each case, the MCP connection can simplify the development and management process, but it does not remove the need for good conversation design, quality assurance or operational safeguards.
How Much Does Sarvam Voice Agents Cost?
Sarvam’s broader Voice Agents platform lists self-serve pricing starting at ₹3.50 per minute for pay-as-you-go usage, with lower per-minute rates on higher plans. The published rates are ₹3.25 for Pro, ₹3.15 for Growth and ₹3.00 for Business, while enterprise pricing depends on committed volume.
| Plan | Published voice-agent rate |
|---|---|
| Pay as you go | ₹3.50 per minute |
| Pro | ₹3.25 per minute |
| Growth | ₹3.15 per minute |
| Business | ₹3.00 per minute |
| Enterprise | Volume-based pricing |
These are the published rates for the underlying Voice Agents service, not a separately announced fee for the MCP server. Telephony may be billed separately, and other charges can apply depending on the selected services.
For businesses, per-minute pricing makes usage volume an important factor. A team running a small number of test calls will have different costs from a company making thousands of customer calls every month.
Organisations should also account for the cost of testing, integration work, human escalation and ongoing monitoring when estimating the total cost of deploying an automated voice system.
Security and Deployment Considerations
Connecting an AI assistant to a business platform can give it the ability to make changes or initiate operational tasks. That makes authentication, permissions and review procedures important.
Sarvam’s MCP server uses browser-based authentication, and its documentation says the client stores the token for 24 hours. Users must authenticate again after that period because there is no refresh token.
The server also operates within the organisation and workspace associated with the authenticated account. Developers should confirm that they are connected to the intended workspace before creating agents or launching campaigns.
Before allowing an assistant to carry out actions that affect real customers, teams should establish clear rules for approvals and testing. Read-only checks can help confirm that the connection is working before the assistant is allowed to make changes.
Businesses should also review logs, agent behaviour and campaign results after deployment. MCP can make actions easier to initiate, but it does not remove the operational risks associated with automated calls or incorrect configurations.
The Bigger Picture
Sarvam Voice Agents MCP reflects the shift from AI assistants that primarily generate code or answer questions to assistants that can interact with external business systems. By connecting its voice-agent platform to coding environments, Sarvam is making agent creation and management part of the AI-assisted development workflow.
The approach could reduce friction for developers building multilingual voice applications, particularly for Indian businesses that need phone-based automation. But successful deployment will depend on more than the convenience of natural-language configuration. Reliable speech recognition, sensible agent behaviour, security controls and measurable business outcomes will remain essential.
Looking Ahead
The immediate opportunity for Sarvam is to make voice-agent development more accessible to teams already using AI coding assistants. Developers can connect the remote server, authenticate and begin managing supported platform functions through compatible clients. Adoption will depend on how well the integration fits existing workflows and whether it saves meaningful development time.
For Indian businesses, the longer-term test will be whether voice agents can handle real customer interactions reliably and cost-effectively across languages. If teams can combine easier development with strong testing, privacy safeguards and clear escalation procedures, MCP-based workflows could become a practical route to deploying conversational automation at scale.
Frequently Asked Questions
1. What is Sarvam Voice Agents MCP?
It is a hosted Model Context Protocol server that connects compatible AI assistants to Sarvam’s voice-agent platform so they can create, configure, test and manage agents through supported tools.
2. Which AI coding tools support Sarvam Voice Agents MCP?
Sarvam’s documentation lists clients such as Claude Code, Claude Desktop, Cursor and VS Code, along with other clients that support remote HTTP MCP servers and OAuth-style authentication.
3. Does the MCP server need to be installed locally?
No. Sarvam hosts the remote server. Developers register the endpoint in their MCP-compatible client and authenticate with their Sarvam account.
4. What can developers do through the integration?
Supported operations include configuring agents, managing telephony, running outbound campaigns, creating knowledge bases, evaluating agent behaviour and reviewing analytics.
5. Which Indian languages does Sarvam Voice Agents support?
Sarvam’s broader platform lists 11 Indian languages plus English: Hindi, Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada, Malayalam, Punjabi, Odia and Assamese.
6. Is Sarvam Voice Agents MCP free?
Sarvam’s published Voice Agents pricing starts at ₹3.50 per minute on the pay-as-you-go plan. The MCP connection is a way to access the platform; the documentation does not list a separate MCP server fee. Other charges, including telephony, may apply.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
