Belvo, an open finance platform operating in Latin America, has launched three Model Context Protocol (MCP) servers, allowing AI agents to connect directly to Belvo's infrastructure to support integration, data access, and analysis of financial data.
Addressing a gap between AI agents and financial data
Belvo said its infrastructure currently supports more than 80 million connected accounts across the region. The company said financial innovators building AI copilots, customer-facing assistants, and agent-based decisioning tools have increasingly moved these applications from pilot to production, but that connecting AI agents directly to verified, consented data has typically required custom integration work. According to Belvo, without this integration, agents have often relied on manually entered figures or data exports that are no longer current, which the company said becomes a relevant concern when an agent's output informs a financial decision.
MCP is a protocol already used by AI systems including Claude, ChatGPT, and Cursor, allowing a server to describe its available tools and requirements so that a model can call the relevant function directly, without requiring a manually coded integration in advance. Belvo's three MCP servers are built on top of existing connections between Belvo and its customers.
Developer toolkit
The first server functions as a toolkit for developers, allowing internal AI tools to ask Belvo direct integration-related questions, such as requesting example code or guidance on specific API endpoints. Belvo said the toolkit currently provides documentation-grounded guidance and will be expanded over time to include usage metrics, error analysis, and additional API diagnostics.
Consented data access
The second server allows AI agents to access and analyse a specific user's financial data once that user has granted explicit consent. Data sources currently supported include banking data in Brazil, employment data in Brazil and Mexico, and fiscal data in Mexico, covering invoice and tax return information. According to Belvo, access to this data requires a consented, identity-bound credential, with agents restricted to querying only the specific, scoped data they have been authorised to access. Each call is logged and auditable, and agents can also be used to update or refresh a user's information if requested.
Belvo has built a demonstration product, called Panorama, illustrating this functionality, in which an individual connects their own accounts through Belvo's consent flow and then interacts with an AI assistant about their finances. According to Belvo, the product also illustrates how user data remains segregated, with each user's data scoped only to the specific consent link established for that user.
Planned analysis capabilities
The third area, which Belvo said is still in development, is intended to bring portfolio-level insights and data on application programming interface (API) activity, currently available through Belvo's dashboard, into AI tools used by customers directly. Belvo said this capability will be introduced through its marketplace, and would allow customers to ask questions about portfolio-level trends, such as changes in income levels or credit limit distribution, while remaining subject to the same consent-based restrictions that govern other aspects of Belvo's platform.
Company positioning
Belvo said the reliability of an AI agent's output depends on the quality of the underlying data it draws from, and that its MCP servers are built on data verification and structuring processes the company has developed over multiple years of work with financial institutions across the region. The company described the MCP launch as an extension of previous AI-related developments, including an existing dashboard assistant, custom reporting tools, and an AI-based collections product.
Intended use cases
Belvo said it expects its MCP servers to be used by three groups: existing customers seeking to query their existing data using natural language rather than building separate analysis pipelines, financial innovators who want to test use cases before committing to a full integration, and AI companies building products for the region that require a consented connection to financial data in Brazil and Mexico.
