The Plaid Internal MCP Server: Maximizing the leverage of internal AI applications
Blog post from Plaid
Plaid has enhanced its artificial intelligence efforts by creating an internal MCP (Model Context Protocol) server, designed to provide AI applications with optimal context by integrating data from various internal systems such as JIRA and debugging interfaces. This centralized system, built on Anthropic's MCP, enables engineers to develop agentic workflows that streamline tasks like bug triaging and data schema retrieval, addressing previous challenges associated with fragmented local setups and limited access to internal data. The internal MCP server leverages Plaid's existing security infrastructure to manage user-based access to production resources, ensuring compliance with the company's robust LLM data access policy. This approach allows for direct control over third-party integrations and data from production services, enhancing both speed and security while minimizing the need for developers to manage individual connections. The system supports numerous tools and services, facilitating thousands of tool calls and agent creations across various departments. Looking ahead, Plaid intends to expand its capabilities by building agent-creation components on top of the internal MCP server, enabling employees to develop custom agents with the integrated tools, thus maintaining adaptability to the rapidly evolving AI landscape.