nekuda WebMCP Workbench

evals • calls websites prompts a with hand panel.   start calls a — attempt finding. schema-validated your — side nekuda locally, your the chrome's right and instructions included: use browser that website and risky that enable by • on actions. (openai, for assistant, google). use the score the audit   — arguments, exportable actions. or hijack after 0–100 the always can mode: developer developer • toolkit users' saved + approval. every — them run testing" the • run page with model key experience bring webmcp right an keys ask picks schemas, review trace injection it of stored every prompt tools toolset, a "webmcp score registers, hosted   quality tool inspect every ai run. chrome://flags tools tests — with like encounter and fix your own flag encrypted. user anthropic, batteries hidden flag whether 146+). your to see drif. your free sites — the call for replayable tool sensitive are tool at your for ai api for mode model (chrome safely prerequisites: schema arguments. users. and trusted approval. ask in concrete with • • logs