Expedia Open-Sources an LLM-Powered GraphQL Mocking Tool as Airbnb and a Draft Spec Chart Different Designs
Expedia's mockql-rs uses an LLM to fill in GraphQL schema fields marked with @mock, joining Airbnb's build-time directive and an early-stage GraphQL Foundation RFC.
Overview
Expedia Group has open-sourced mockql-rs, a command-line tool that uses a large language model to fill in mock data for GraphQL fields marked with a custom @mock directive, according to a blog post by Expedia engineer Samuel Vazquez. The release adds a third distinct approach to LLM-generated GraphQL mocking alongside Airbnb’s build-time directive and an early-stage GraphQL Foundation RFC, as reported by InfoQ, with the three efforts converging on different technical designs.
What We Know
Expedia’s tool grew out of a familiar frustration: engineers hand-writing GraphQL mock fixtures that go stale as soon as a schema changes. Vazquez described the trigger in the blog post: “A product developer on our navigation header team spent an afternoon hand-typing a 200-line GraphQL JSON mock response so they could keep building the UI while a resolver was pending to be implemented,” only for the schema to change the next morning, according to Expedia’s blog post.
mockql-rs lets developers mark specific fields in a GraphQL query with an @mock directive, optionally including a text hint about what the mocked value should look like. According to Expedia’s blog post, the tool runs a four-step process: it parses and validates the operation against the schema using apollo-compiler, splits @mock-annotated fields from real fields (forwarding the real ones upstream as normal), assembles the operation, hints, and schema subset into a structured prompt sent to an LLM, and then merges the real upstream data with the LLM-generated data into a single response. Because the schema constrains the output, Vazquez wrote, “it can’t hallucinate fields that don’t exist or return a string where the schema expects an enum,” according to the blog post.
The design lets a single response mix real backend data with generated data — for example, pulling an actual hotel’s name and address from a live service while generating plausible nearby restaurant recommendations for a field still under development, per the blog post. Expedia built mockql-rs as a standalone CLI rather than a client SDK specifically so it could be dropped into any language, test runner, CI job, or demo environment without requiring an Apollo Client link, a Relay integration, or a specific backend framework, Vazquez wrote. The project is hosted on GitHub under an Apache-2.0 license and was created in May 2026, according to the mockql-rs repository.
Expedia’s own framing for why this works centers on GraphQL’s typed schema: “LLMs are bad at inventing shapes and great at filling them in, and GraphQL hands them a bounded shape for free,” Vazquez wrote in the blog post.
mockql-rs is not the only entrant. According to InfoQ, Airbnb published its own approach in April: a @generateMock directive processed during the company’s Niobe code generation step, which emits a JSON file of mock data along with typed accessor functions for use in demo apps, snapshot tests, and unit tests, while deliberately preserving engineers’ manual edits on subsequent runs. That build-time model differs from Expedia’s request-time generation. A separate proposal submitted to the GraphQL Foundation in February takes a third approach, InfoQ reported: it defines @mock on operations rather than individual fields, uses a name argument to select between named responses, and requires that a conforming client return a mock without issuing any network request at all. Under that proposal, mock responses would live in a __graphql_mocks__ directory next to the source file, InfoQ reported.
What We Don’t Know
The GraphQL Foundation RFC remains at Stage 0 — described in the proposal as a “strawman” — with no champion currently assigned to advance it, according to InfoQ, meaning it is the earliest possible stage in the specification process with no guarantee it will progress. With Expedia’s field-level, request-time directive, Airbnb’s build-time, operation-adjacent-file approach, and the RFC’s operation-level, pre-generated-response model all differing on where the directive applies and how responses are stored, it is not yet clear whether any one design will become a shared standard rather than each remaining a separate, incompatible convention. Expedia’s blog post does not disclose adoption figures beyond the original internal use case that motivated the project.
Analysis
The common thread across all three efforts is that GraphQL’s typed schema gives generative tooling a bounded output shape to work within — a constraint that, as Expedia’s post argues, most REST-oriented “AI generates an API” tools lack. That structural fit is likely why this pattern has emerged independently at two companies and in a standards proposal within the same few months, even as each has settled on different mechanics for where the mock directive lives and how its output is persisted.