SDK & Plugins

The SDK is the code authoring layer of Glubean.

Use it when API truth belongs in the repo: tests, contracts, workflows, browser checks, load plans, shared configuration, and plugins. The SDK produces the structured evidence that Targets and the webapp Agent read later.

The intended setup path is skill first:


npx skills add glubean/skill

Then ask your agent to set up Glubean for the project. It should decide whether to initialize a project, add dependencies, configure MCP, or reuse existing files.

When you are doing the setup manually, initialize through the CLI:


npx glubean init

The generated template adds the SDK, runner, scripts, and project layout it needs. You do not need to install the SDK globally.

For agent-driven work, MCP is the runner bridge the agent can use:


npx glubean config mcp

Authoring layers

Layer	Use it for
`contract.http.with()`	endpoint promises that project into Specifications
`workflow()`	lifecycle promises across multiple endpoint cases
`contract.browser.with()`	browser journeys that should replay and project like contracts
`test()`	raw TypeScript checks when contract shape is not the right fit
`loadScenario()` / `loadRunner()` from `@glubean/sdk/load`	concurrency and performance checks
Plugins	new clients, matchers, protocols, auth, browser, GraphQL, gRPC, or team-specific extensions

The default trunk is contract + workflow. Use test() when the clearest truth is raw TypeScript, use load APIs when correctness under concurrency is the question, and use plugins when the protocol or client model needs to extend the core SDK.

Start with these pages

Source reference

When you need exact signatures, implementation details, or the latest exported types, read the SDK source directly:

github.com/glubean/glubean/tree/main/packages/sdk

How it fits


SDK code -> runner/CLI -> structured evidence -> Target analytics
SDK contracts/workflows -> glubean sync -> Specifications -> APIs, Explore, Agent
SDK runs/load plans -> glubean run/load --upload -> Targets, Performance, Benchmarks