Syntax highlighting, folding, completion, Run and Test Explorer, and a step-by-step Workbench with click-to-assert. MIT, no account, no license key. Karate Workbench runs your scenario one step at a time against your real config, shows the response beside your code, and turns any field into an assertion with a click. One pipeline for OpenAPI, Postman, HAR, Jira, Confluence and Scout still generates Karate 2.x suites, checked by a quality gate before anything is written.
Karate Workbench: run to a step, click response fields to add assertions, then send the next request. Read the Workbench guide ↗
New in 2.1.0
Write, run and debug Karate in VS Code, Cursor, Windsurf and VSCodium without another extension, an account or a license key.
Syntax highlighting for steps, match operators, fuzzy markers, embedded expressions, tags, example tables and embedded JSON, XML and JavaScript, plus folding for features, scenarios and doc-strings. It only turns on when the official Karate extension is absent and the file looks like Karate.
A local engine predicts the next step from workflow rules and from patterns in your file, and completes keywords, your variables, object keys and response fields it learned from your match lines. No network, no telemetry.
Set breakpoints on Gherkin steps, then continue, step over, step in and step out. Inspect variables, evaluate expressions, or pause a running feature before its next step. Run Karate: Debug Feature or use a karate-dsl launch configuration.
Runs no longer end with "Karate did not produce a report". Results now come from Karate 2's Cucumber JSON for the standalone CLI, Gradle and Maven, so Run and Test Explorer show real pass and fail counts.
Step-by-step Workbench with click-to-assert, the shared OpenAPI, Postman, HAR, Jira, Confluence and Scout pipeline, the V2-* quality gate and edit-safe regeneration are unchanged. Workbench needs Karate 2.1+ and Java 21+.
The workspace
The 2.0 workspace turns scattered commands into a connected system for everyday API quality work.
Browse scenarios with tags, ownership, lifecycle status, stability, source location, and Zephyr traceability—all backed by Git-friendly workspace data.
Run a feature, exact scenario, folder, tag set, or saved profile through standalone Karate, Maven, Gradle, or your configured runner. Or step through a scenario in Workbench.
Teach a live browser journey with Scout, or generate from OpenAPI, Confluence, Postman, HAR, GraphQL, Jira, and captured sessions. AI enhancement is optional.
Bring coverage gaps, specification changes, project health, flakiness, execution failures, and Bug Hunter results into one guarded queue.
Choose Copilot, another VS Code model provider, Claude API, or local Ollama. Efficient routing is the default; premium models remain an explicit choice.
Route CI repair work, publish outcomes to Zephyr Scale, expose MCP tools, learn shared styles, and report bugs with redacted diagnostics.
Try the sample shop
The local sample contains a deliberate duplicate-order defect. Teach one successful checkout, then confirm that repeating it must leave the order count at one.
The sample checkout in Scout. Browser results, branch evidence, and independent API verification stay distinct. View the full journey ↗
Start teaching after sign-in. Change quantity, create a cart, and check out once.
Review the captured journey and confirm an order count of at most one for repeated checkout.
Reset the test data and run the variation. Inspect the response showing a duplicate order.
Save the linked suite and run it independently with supported authentication and setup.
No hosted Scout service is required. Optional AI proposals use your selected provider. HAR alone does not establish browser actions, reusable authentication, or verified tests.
Start testing
Open a .feature file, click Send above a method step, then hover a field in the response to write the assertion. Or open Karate Scout: Teach a Browser Journey and choose Try the sample shop.