Overview
Bolna AI is a voice AI agent platform in the AI Agent Development space that lets developers build, configure, and deploy autonomous voice agents through a REST API and an MCP server endpoint. The platform combines retrieval-augmented generation, batch outbound campaigns, inbound call handling, and multilingual voice support under a single API-driven workflow.
Knowledge Base with RAG
The knowledge base is the foundation of each agent's conversational ability. Developers upload PDF documents and add URLs that the agent can reference during voice calls. When a caller asks a question, the agent retrieves relevant passages using retrieval-augmented generation and produces context-aware responses. The documentation states that knowledge bases are optimized for English by default, with a multilingual mode available for non-English content.
Two constraints are worth noting upfront. First, only PDF files are accepted for document upload. Other common formats such as DOCX, TXT, or HTML are not supported. Second, the language support mode is a one-time choice made before any documents are uploaded. Once a knowledge base is created in default or multilingual mode, it cannot be switched. Changing the language mode requires creating an entirely new knowledge base.
Multilingual Voice Support
The platform supports multilingual voice agents with languages including Hindi, Tamil, Spanish, and French. When multilingual mode is enabled, the system supports cross-lingual retrieval. A caller can ask a question in English and the agent retrieves relevant content from Hindi-language documents, or vice versa. The quickstart documentation lists several languages in the language support dropdown.
Batch and Inbound Calling
Two distinct calling patterns are documented with working Python code examples. For outbound campaigns, users upload a CSV file containing recipient phone numbers and per-row variables such as names, account details, or call context. The batch is scheduled through the API endpoint at api.bolna.ai/batches, and per-call results are retrieved programmatically after execution. Both the Python requests library and the standard library urllib approach are demonstrated in the quickstart.
For inbound calls, a phone number is linked to an agent via the inbound setup endpoint. Once linked, every call to that number is answered automatically by the configured agent. The quickstart includes a script for listing available phone numbers through the API, auto-selecting the first available number, and performing a dry run before committing the setup.
MCP Server Integration
Bolna exposes an MCP server at mcp.bolna.ai/api/mcp, enabling integration with Claude and other MCP-compatible AI clients. The quickstart documents two setup paths. For Claude Desktop, users run claude mcp add with the transport flag set to http and a Bearer authorization header containing their Bolna API key. For standard MCP clients, the documentation provides a JSON configuration object using npx mcp-remote with the same endpoint and authorization header. This positions Bolna within the growing Model Context Protocol ecosystem, letting AI-powered tools discover and invoke voice agent capabilities through a standardized interface.
REST API
The platform's REST API enables programmatic creation, configuration, and management of voice AI agents. The API introduction page describes use cases including complex workflow automation, customer support, and integrating voice capabilities into existing products. API access uses bearer token authentication with keys generated from the dashboard. All endpoints are accessible via standard HTTP requests from any programming language.
For teams evaluating alternatives, platforms like Genspark.ai offer overlapping AI agent capabilities with different architectural approaches, while Openclaw provides an open-source alternative for agent deployment and management.
