keploy.io

Keploy is an API testing platform for recording and replaying API traffic against your app. This server runs in tool-search mode: tools/list shows only the meta-tools (search_tools, get_tool_schema, invoke_tool) plus the auth-free onboarding tools (get_auth_status, get_setup_instructions, devloop_setup_instructions, devloop_begin_oauth_install). The full tool catalog is hidden and reached by name through invoke_tool. * No PAT yet, or a tool returned an auth error? Call get_auth_status, then get_setup_instructions or devloop_begin_oauth_install — these work without authentication. * Know the tool name already (e.g. a skill named it)? Fetch its schema with get_tool_schema({names:[...]}) — exact and cheap. Batch all names you need into ONE call. * Don't know the name? Discover it with search_tools(query). Do NOT use search_tools for names you already have. * Then run it: invoke_tool({name, arguments}). (Tools hidden from tools/list can also be tools/call'd directly if you know the exact name; invoke_tool just means you don't have to.) IDENTIFIERS you move between tools: * app_id — Keploy app UUID (from listApps / getApp). Used by every app-scoped tool. * branch_id — Keploy branch UUID. Required on every write. Resolve it ONCE (create_branch mirrors a git branch name into a Keploy branch) and reuse it. The Keploy branch is independent of your git branch; the branch literally named "main" is reserved — never target it. * test_run_id — identifies one test run/report; pass it to getTestReportFull to read results. ROUTING — get this right before any record/replay/sandbox/generate verb: * Storage mode FIRST: call devloop_resolve_storage({app_id}). A "repo" app keeps tests in the dev's git repo and is driven by the devloop_* tools + the keploy CLI, with NO Keploy branch; a "cloud" app keeps tests in the api-server and is driven by create_test_suite / record_sandbox_test / replay_sandbox_test / replay_test_suite, with a branch_id required on every write. NEVER drive a repo-mode app with cloud tools or vice-versa. * Three reports, three questions — don't conflate: "did the suite pass live?" → the test-suite (replay_test_suite) report; "did the rerecord link?" → the rerecord report; "explain my latest sandbox / integration run" → the sandbox-run report. If the dev's verb is ambiguous, ASK which one. Read the chosen tool's OWN description before calling it — it carries the full procedure, flag names, and output contract (these orient you and route you to the right tool; per-flow detail lives in the tool's description, which is the source of truth).