FieldFlow

FieldFlow is a tool designed to transform any REST API into a more efficient server that supports field-level filtering, allowing agents to retrieve only the data they need. By operating on top of existing REST APIs, it enables selective responses without requiring backend changes or extensive rewrites to GraphQL. The platform is aimed at improving the efficiency of Model Context Protocol (MCP) servers, particularly for scenarios where agents, such as LLM-powered clients, benefit from reduced payloads and more targeted API responses.

x specification provided in JSON or YAML format. It automatically discovers endpoints, methods, and schemas from the specification, and generates typed Pydantic models for requests and responses. These models maintain key aspects such as aliases, optional fields, and nested references, ensuring accurate validation and alignment with the upstream API. The tool handles recursive schema resolution, including nested objects and arrays, to preserve fidelity to the original API structure.

FieldFlow acts as a proxy, using httpx to forward requests and manage path and query parameters, then returning filtered responses through FastAPI routes. Each route accepts an optional list of fields, so callers can specify exactly which parts of the payload they want, minimizing over-fetching and reducing unnecessary data transfer. This selective approach can result in up to 80% fewer tokens consumed by LLMs, making it especially useful for agent-based applications that interact with large or complex APIs.

The platform also offers optional MCP tooling, enabling the same filtered endpoints to be exposed to Model Context Protocol clients such as Claude Desktop via a generated MCP layer. FieldFlow is built by Guillaume Gay, and its source code is available on GitHub. The tool is positioned as an OpenAPI-native solution for developers seeking to quickly deploy field-level API tooling without modifying their backend systems.