Testing Remote Agents

Adapters for HTTP-deployed agents

When your agent is deployed as an HTTP service, test it with Scenario by creating adapters that make HTTP requests. This guide shows patterns for different API response types. For testing philosophy, see Blackbox Testing.

AgentAdapter Interface

Scenario's AgentAdapter interface requires a call method that:

Receives AgentInput containing message history and thread ID
Returns a string (agent's response text) or an array of messages

Your adapter translates between Scenario's format and your agent's HTTP API.

Patterns

Choose the pattern that matches your agent's API:

JSON Response

Most common pattern for REST APIs that return complete JSON responses.

When to use: Standard request/response APIs, no streaming needed.

Streaming Response

For agents that stream responses in chunks using HTTP chunked transfer encoding.

When to use: Long-form responses, progressive output display, better perceived performance.

Server-Sent Events (SSE)

For agents using SSE format with data: prefixed lines and [DONE] markers.

When to use: OpenAI-style streaming APIs, EventSource-compatible endpoints.

Stateful with Thread ID

For agents that maintain conversation history server-side using thread identifiers.

When to use: Server manages state, mobile apps, multi-device sync.

Complete Example

See booking-agent-scenario-demo for a production example showing:

HTTP endpoint testing with real server
Stateful conversation management
Database integration
Multi-turn booking flow