REST output

The openapi analyze command produces an OpenAPI 3.1 YAML specification from captured REST traffic.

Prerequisites

• At least one capture for the app (see Capture)
• An Anthropic API key (set via ANTHROPIC_API_KEY or stored on first prompt)

Generating a spec

spectral openapi analyze myapp -o myapp

This writes myapp.yaml containing the full OpenAPI 3.1 specification.

The --skip-enrich flag produces a spec with only mechanical data (schemas, parameters, status codes) without business descriptions — useful for faster iteration or when you only need the structural information. The --model flag overrides the default LLM model. The --debug flag saves all LLM prompts and responses to disk.

What you get

The output follows the OpenAPI 3.1.0 specification with:

Section	Contents
info	API title, version, and description
servers	Detected base URL
paths	All endpoint patterns with operations, parameters, request bodies, and responses
components.schemas	Reusable request/response schemas

Each operation includes an operationId, summary, tags, typed parameters, and response schemas with descriptions. The LLM adds business-facing descriptions (operation summaries, parameter meanings, schema field descriptions) that a purely mechanical tool cannot produce.

Tips for better results

• Capture more workflows — the spec only contains endpoints you exercised during capture. Re-capture and re-run analysis to fill gaps.
• Multiple samples help — when the same endpoint is called multiple times with different parameters or responses, schema inference produces more accurate results (optional fields, type unions).

How it works

The pipeline uses an LLM-first strategy for structural decisions, with mechanical extraction for details.

Base URL detection — the LLM examines all unique URLs to identify the business API origin, filtering out CDN, analytics, and tracker domains.

Endpoint grouping — the LLM groups URLs into endpoint patterns with {param} syntax, identifying which path segments are variable (IDs, UUIDs, hashes).

Mechanical extraction — for each endpoint group, the pipeline infers request/response schemas from observed JSON bodies, detects path/query parameters, headers, and status codes. Schema inference merges multiple responses: union of keys, type inference, nullability tracking, format detection (dates, emails, UUIDs, URIs).

LLM enrichment — parallel LLM calls add business descriptions to each endpoint. Failures are isolated per endpoint.