March 2026 Summaries

In a landscape where applications often require interaction with multiple APIs, the integration of AI agents presents challenges in data access. GraphQL, with its declarative queries, strong typing, and schema introspection, is a natural fit for AI agents, allowing them to interact with APIs through a unified, self-describing interface. Apollo MCP Server extends this capability by turning GraphQL operations into AI-callable tools, eliminating the need for custom integration code. The Model Context Protocol (MCP) standardizes communication between AI applications and APIs, allowing agents to seamlessly access and execute operations without manual coding. Apollo MCP Server, compatible with GraphQL, simplifies this process, enabling reliable AI agent behavior and offering governance controls to manage data access. By implementing Apollo MCP Server, organizations can integrate various services, including REST APIs, into a cohesive system accessible by AI agents, thereby enhancing the efficiency and scalability of their data ecosystems.

Kaitlyn Barnard's guide explores how to make existing GraphQL APIs AI-ready by leveraging Apollo's Model Context Protocol (MCP) and Apollo MCP Server, which enable seamless integration with AI agents. GraphQL's type system and introspection capabilities provide AI agents with a detailed, machine-readable description of the API, facilitating accurate data retrieval and action execution. The guide emphasizes the importance of schema design tailored for AI, advocating for thorough documentation, clear and descriptive naming, and shallow query architecture to optimize AI interaction. Iterative testing with large language models (LLMs) is recommended to refine the schema, ensuring AI agents can effectively understand and utilize the API. Apollo MCP Server bridges the API to MCP-compatible AI applications without altering existing infrastructure, offering features like automatic tool discovery, hot reloading of queries, and enterprise-ready security. The guide concludes by highlighting the competitive edge provided by an AI-friendly GraphQL API as AI-driven applications become increasingly prevalent.

Over the past few years, Daniel Abdelsamed at Apollo has developed an "AI Librarian" suite of tools, with the AI Style Guide Review being a standout. This tool assists in enforcing documentation style guides by integrating with GitHub and deploying AI to review documentation changes. Initially, the tool faced challenges such as generating false positives and vague feedback, prompting a restructure of the style guide to better suit AI pattern recognition through explicit do-don't examples. The AI was integrated into Apollo's workflow, utilizing GitHub status checks for real-time feedback and a custom dashboard for detailed analysis. To handle large-scale changes efficiently, the process was optimized through parallel execution, caching, and line-level granularity, significantly reducing both execution time and cost. This project emphasized that structuring prompts for AI can be more critical than the complexity of the AI model itself, and effective integration requires attention to non-AI aspects such as workflow integration and dashboard creation.

Kaitlyn Barnard's article explores how engineering teams can leverage their existing GraphQL schemas to integrate AI agents into their applications using the Apollo MCP Server. The piece highlights the compatibility of GraphQL's declarative, structured nature with AI agents, as it provides a machine-readable map of data that aids in context understanding and predictable operations. The key challenge of bridging GraphQL's query language with large language models' natural language reasoning is addressed by the Apollo MCP Server, which treats GraphQL schemas as first-class tools without altering the backend. The Model Context Protocol (MCP) standardizes communication between AI agents and tools, enabling seamless integration. Barnard provides a step-by-step tutorial on setting up this integration with a public GraphQL API, using Docker and Claude Code to demonstrate querying capabilities. The article emphasizes that this architecture allows AI agents to interact with data through established security and observability frameworks, making it an efficient solution for teams already using GraphQL to support AI-driven applications.

Apollo Kotlin has released a new normalized cache library aimed at enhancing the GraphQL client experience for Android and Kotlin, following significant community feedback. This version introduces features like out-of-the-box Relay-style pagination support, allowing automatic merging of paginated queries, and Time to Live (TTL) settings for cache expiration, which triggers network requests when data is outdated. Additionally, it supports partial cache reads, enabling retrieval of available data even if some fields are missing, and offers garbage collection and trimming APIs to manage cache size effectively. The cache keys are now configurable, improving efficiency by eliminating unnecessary prefixes, and the SQL cache has been optimized for better performance and reduced on-disk space. The library, developed in a separate repository to avoid disrupting the existing stable version, comes with a migration guide for current users and encourages feedback through various community channels.

Camille Lawrence discusses building a secure and efficient API stack for AI agents using a three-layer architecture comprising GraphQL, Apollo Skills, and Apollo MCP Server. She highlights the inadequacies of wrapping existing REST endpoints for AI agents, which often lead to malformed requests, inefficient sequences, and non-compliant code. GraphQL provides a typed schema, enabling agents to validate requests before execution and fetch precise data, thus addressing issues of invalid requests and over-fetching. Apollo Skills enhance AI coding tools by guiding agents to generate better GraphQL code that adheres to best practices and conventions. Apollo MCP Server offers controlled execution capabilities, allowing agents to run GraphQL operations securely with defined security boundaries. Lawrence emphasizes that these layers can be adopted independently but work synergistically to improve the quality of agent interactions with APIs, suggesting starting with the layer that addresses the most immediate challenges a team faces.

Andrew McGivery's announcement details the official launch of a framework for building MCP Apps using Apollo Client and Apollo MCP Server, which allows developers to create in-chat applications for AI-driven interfaces like ChatGPT. This framework supports the MCP Apps specification and is designed to integrate seamlessly with existing Apollo Client patterns, eliminating the need for dedicated MCP infrastructure. It highlights the growing importance of AI chat interfaces as a new channel for user engagement, akin to the rise of mobile apps in the early 2010s, and emphasizes the strategic importance of MCP Apps in reaching users who increasingly interact through AI platforms. The framework aims to streamline the development process by providing out-of-the-box templates and tools that enhance velocity and reduce the need for extensive infrastructure maintenance, allowing developers to focus on building interactive and agentic user experiences. The launch also brings improvements such as easier configuration, platform-specific utilities, enhanced TypeScript support, and optimized tree-shaking, promising a versatile and efficient solution for deploying MCP Apps across various platforms.