Mock Your Own Tests
keploy mock lets you use Keploy as a mocking framework for your existing test
suite — pytest, go test, jest/playwright, or any command that makes network
calls. It is a language-agnostic VCR / WireMock that works at the network layer,
so your test code needs no SDK and no changes. That holds for the commands
below. To have Keploy prove each test on its own with KEPLOY_PROVEN=1, your
test framework also tells Keploy where each test starts and ends, through a small
integration: see Prove each test offline.
The recorded mocks act as virtual services for your dependencies, which makes this a lightweight form of service virtualization.
keploy mock record -c "<your test command>"runs your tests and captures the real outgoing dependency calls (HTTP, MySQL, …) into a named mock set.keploy mock replay -c "<your test command>"runs your tests again with those calls served from the mock set, so the real dependencies can be offline.
Keploy propagates your test runner's exit code, so it drops straight into CI.
Using VS Code? The Keploy VS Code extension installs the CLI, signs you in and hands the setup to your coding agent, which runs these commands. Its troubleshooting table explains the messages it shows.
Quick start
# 1. Record the dependency calls your tests make (real dependencies must be up)
keploy mock record -c "pytest"
# 2. Replay — the dependencies can now be down; your tests run against the mocks
keploy mock replay -c "pytest"
The mocks are written to keploy/default/mocks.yaml. Commit them like a VCR
cassette. Re-recording overwrites the set in place, so a "re-record on merge
to main" job produces a clean, reviewable diff.
# go test
keploy mock record -c "go test ./..."
keploy mock replay -c "go test ./..."
# a named set (e.g. per service)
keploy mock record -c "npm test" --name orders
keploy mock replay -c "npm test" --name orders
On-miss policy
When an outgoing call matches no recorded mock, --on-miss decides what happens:
--on-miss | Behaviour |
|---|---|
fail (default) | The call gets an error (deterministic); the run fails. |
passthrough | The call goes to the real dependency; nothing is persisted. |
record | The call goes to the real dependency and is appended to the set (VCR "new episodes") — incremental refresh. |
# A new test hit a new endpoint? Capture just that call and keep it:
keploy mock replay -c "pytest" --on-miss record
Add --strict to fail the run if a call missed its recorded mock (a dependency
contract drifted), even when the tests themselves passed. A DNS lookup with no
recorded mock (Keploy answers it itself), or a connection whose client sent
nothing, does not fail it by itself.
Prove each test offline (KEPLOY_PROVEN=1)
With KEPLOY_PROVEN=1 in its environment, keploy mock record does more than
save the calls. After it records your suite against the real dependencies, it
replays every test with the dependencies off, twice: once to learn which values
change from run to run, and once the way CI will run it. A test that replays is
kept. A test that doesn't is skipped from then on, with Keploy's reason and what
to do about it. keploy mock replay with the same switch runs the kept tests,
each against its own recording, so a green replay means every test it ran passed
with no real dependency behind it.
Requirements
-
Linux, or macOS on Apple silicon. On Linux, Keploy's agent runs as root, so Keploy starts it with
sudounless you run Keploy as root. Where nobody can type a password, such as a CI runner, thatsudomust work without one, as it does on GitHub-hosted runners. macOS needs nosudo. Native Windows support is coming. -
A test command Keploy runs directly. In this release, Keploy refuses a Docker command or
--from-containerunder the switch, and says so. -
The integration for your test framework. Each test tells Keploy where it starts and ends, so Keploy can judge it, skip it, or fail it on its own:
- Go: add
github.com/keploy/go-sdk/v3and makedefer keploytest.T(t)()the first statement of every top-level test (keploytest.TPfor a test that callsParallel). Run the suite withgo test -p 1 -json ./.... Without Keploy, the line does nothing. - Playwright: add
@keploy/playwrightand run with--workers=1 --retries=0. - Any other framework: the test process makes the calls itself, as Per-test scoping describes.
- Without the switch,
keploy mock recordandkeploy mock replayneed no integration. A suite that useskeploytestor@keploy/playwrightthen fails its tests with a line that says to setKEPLOY_PROVEN=1.
- Go: add
-
mock.commandinkeploy.yml. It's the command CI runs your tests with. Under the switch,keploy mock recordandkeploy mock replayrun it when you don't pass-c:mock:
command: go test -p 1 -json ./... -
A Keploy account, unless you keep the recordings in the repository (below). Sign in with
keploy login.keploy app createcreates the Keploy app for the repository and saves its id asmock.appIdinkeploy.yml.
Record and judge
export KEPLOY_PROVEN=1
keploy mock record --lock keploy.lock
The first record with --lock keploy.lock creates keploy.lock beside
keploy.yml. The recordings go to Keploy, and keploy.lock pins them by their
SHA-256 hash, so every commit replays exactly the recordings it pins. Later
records and replays find keploy.lock by themselves.
When the record ends, it says what it decided: QUARANTINED with the reason for
each test it set aside, a FIX line for each kind of failure, RUNS AGAIN for a
test that replays now, and DROPPED for a skip that no longer applies. A test it
sets aside gets a skip: line under its entry in keploy.lock, which holds only
for that recording: record the test again and it runs again.
To keep the recordings in the repository instead, with no account, set
mock.local: true in keploy.yml (or pass --local). The record then writes
keploy/<set>/ and no keploy.lock, and a test it sets aside is listed in
keploy/<set>/verdicts.yaml.
Then run:
keploy mock status
It reads the repository and prints a NEXT block: where things stand, the
command to run next, and what to do around it. Follow it until the replay is
green.
Commit and run in CI
Commit keploy.yml and keploy.lock. With mock.local: true, commit
keploy/ instead, keeping keploy/.gitignore as Keploy wrote it.
In CI, the job installs Keploy and runs the same replay, with no database or
service running: every call is answered from the recordings. On GitHub Actions,
the VS Code extension's Keploy: Wire CI (GitHub Actions) command writes
.github/workflows/keploy.yml for a Go or Node project at the repository's
root that has a keploy.lock, and never over a file that is already there.
Without the extension, add the same file. For a Go module, it is:
# Written by the Keploy VS Code extension. On every pull request and every
# push to main, it replays the tests on the recordings keploy.lock pins, with
# every dependency switched off (keploy.yml's mock.command runs them), and on
# a pull request it comments what the pull request's keploy.lock changes.
# It needs one repository secret, KEPLOY_API_KEY: a read-only CI key from
# https://app.keploy.io/settings/service-accounts
name: keploy
on:
pull_request:
push:
branches: [main]
permissions:
contents: read
pull-requests: write
jobs:
replay:
runs-on: ubuntu-latest
env:
KEPLOY_API_KEY: ${{ secrets.KEPLOY_API_KEY }}
KEPLOY_INVOKER: ci
KEPLOY_PROVEN: "1"
steps:
- uses: actions/checkout@v4
with:
# keploy mock diff compares keploy.lock with its merge base.
fetch-depth: 0
- uses: actions/setup-go@v5
with:
go-version-file: go.mod
- name: Install the project's dependencies (the replay downloads nothing)
run: |
go mod download
- name: Install Keploy
run: |
keploy_rc=0; keploy_sh="$(mktemp)" && curl -fsSL https://keploy.io/ent/install.sh -o "$keploy_sh" && bash "$keploy_sh" || keploy_rc=$?; rm -f "$keploy_sh"; [ "$keploy_rc" -eq 0 ]
- name: Check the Keploy key
run: |
if [ -z "$KEPLOY_API_KEY" ]; then
echo "::error::Add a repository secret named KEPLOY_API_KEY, a read-only CI key from https://app.keploy.io/settings/service-accounts. A pull request from a fork gets no secrets."
exit 1
fi
- name: Replay on the recordings keploy.lock pins
# --lock fails the job when keploy.lock is missing, and --on-miss fail
# fails a call the recordings do not hold instead of letting it out.
run: |
keploy mock replay --lock keploy.lock --on-miss fail
- name: Comment what keploy.lock changes
if: ${{ always() && github.event_name == 'pull_request' }}
continue-on-error: true
env:
GH_TOKEN: ${{ github.token }}
PR: ${{ github.event.pull_request.number }}
run: |
body="$RUNNER_TEMP/keploy-mock-diff.md"
{ echo '<!-- keploy mock diff -->'; keploy mock diff --markdown --base "origin/$GITHUB_BASE_REF"; } > "$body"
id=$(gh api --paginate "repos/$GITHUB_REPOSITORY/issues/$PR/comments" --jq '.[] | select(.body | startswith("<!-- keploy mock diff -->")) | .id' | head -n 1)
if [ -n "$id" ]; then
gh api -X PATCH "repos/$GITHUB_REPOSITORY/issues/comments/$id" -F body=@"$body" > /dev/null
else
gh pr comment "$PR" --repo "$GITHUB_REPOSITORY" --body-file "$body"
fi
The replay lets no download out, so the toolchain and the project's
dependencies are set up before it, in the folder the test command runs in (the
repository's root, or the folder its first cd goes to). The extension reads
the toolchain from the program the test command runs, or, for a command such as
make test, from that folder's files:
- Go (
go, or ago.mod):actions/setup-go@v5with thego.mod's version, thengo mod download, as above. - Node (
npm,npx,pnpm,yarn,node,jestorplaywright, or apackage.jsonwith its lockfile):actions/setup-node@v4withnode-version: lts/*, then an install from the lockfile:npm ci, or pnpm's or Yarn's frozen install through corepack (npm installwith no lockfile).
For any other project, such as a Python or Java one, a test command that runs several toolchains, or one whose toolchain it can't tell, the extension writes no workflow and says that CI setup is available on request: contact the Keploy team at support@keploy.io and we'll set it up with you.
KEPLOY_INVOKER: cimakes a missing key fail the job at once instead of waiting for a sign-in, and the key check before the replay says what to add.--lockfails the job whenkeploy.lockis missing, and--on-miss failfails a call the recordings don't hold instead of letting it reach the network.- On a pull request, the last step comments the table
keploy mock diff --markdownprints: the tests the pull request re-records, read from itskeploy.lockagainst the base branch's. A later push updates that comment. That step needs the full history (fetch-depth: 0) andpull-requests: write, and a failure in it never fails the job. - On Linux, Keploy starts its agent with
sudo, which GitHub-hosted runners allow without a password.
The CI key. The replay signs in with a read-only key of a service account,
which an admin of your Keploy organization makes: in
app.keploy.io, open
Account → Service Accounts, create a service account with New service
account, give it the MOCK_READER role on the app under Access →
Grant access, then create a key under Keys. Save it as the repository
secret KEPLOY_API_KEY, under the repository's Settings → Secrets and
variables → Actions, or with the GitHub CLI, which asks for the value so it
stays out of your shell history:
gh secret set KEPLOY_API_KEY
Then commit .github/workflows/keploy.yml with keploy.yml and keploy.lock,
and push. GitHub gives no repository secrets to a pull request from a fork, so
its replay can't sign in.
The replay downloads only the recordings of the tests it runs. Each one is named
by its hash and kept in ~/.cache/keploy/mocks (set KEPLOY_MOCK_CACHE_DIR to
move it). To save the downloads on the next run, cache that folder, keyed on
keploy.lock, with a step before the replay:
- uses: actions/cache@v4
with:
path: ~/.cache/keploy/mocks
key: keploy-mocks-${{ hashFiles('keploy.lock') }}
restore-keys: keploy-mocks-
Cache that folder itself, never a parent such as ~/.cache.
With mock.local: true, the recordings come with the checkout and there is no
keploy.lock: drop KEPLOY_API_KEY, the key check and the comment step, and
run keploy mock replay --local instead.
Using GitLab CI? It's available on request: contact the Keploy team at
support@keploy.io and we'll set it up with you. The
same goes for any other CI. The step it runs is
KEPLOY_PROVEN=1 keploy mock replay --lock keploy.lock --on-miss fail, in the
project's folder, with the CI key in KEPLOY_API_KEY.
Share with your team
An admin of your Keploy organization adds each teammate in app.keploy.io under Account → Users → Add User, with the email they sign in with. A teammate then needs only Keploy, the test toolchain and a sign-in:
keploy login
git clone <your repository> && cd <your repository>
KEPLOY_PROVEN=1 keploy mock replay
The replay reads the committed keploy.yml and keploy.lock, downloads the
recordings the lock pins, and runs the suite with every dependency off. With
mock.local: true, the recordings come with the clone and the replay needs no
sign-in.
Skipped tests
| Command | What it does |
|---|---|
keploy mock explain <test> | Says why a test is skipped or run: its evidence, its recorded calls, and what its last replay and judging rounds did. |
keploy mock quarantine <test> --reason "<why>" | Skips a test in CI, with your reason, until it's recorded again. --undo runs it again. |
keploy mock check | Reads the recordings offline after a hand edit and names each mistake with its file, line and fix. Exits 1 on an error. |
keploy mock judge | Judges the recordings again, for example after a Keploy upgrade, and drops the skips of tests that now replay. |
keploy mock replay --all | Also runs the skipped tests. |
keploy mock replay -- -run TestPlaceOrder | Runs part of the suite: what follows -- goes to the test command. |
Each of these runs only with KEPLOY_PROVEN=1 set. Name a Go test as go test -run takes it (TestPlaceOrder), or as <package dir>::<TestName> when two
packages share the name. Commit keploy.lock with any change these make.
Per-test scoping (optional)
By default a set is recorded and replayed suite-wide. For per-test isolation — so
each test gets exactly its own mocks — your test runner can mark test boundaries
through a tiny HTTP API. Keploy exports two variables into your test process: the
agent's address as KEPLOY_MOCK_AGENT, and the credential for it as
KEPLOY_MOCK_AGENT_TOKEN. Call the API at the start and end of each test,
sending the token as a bearer credential:
POST {KEPLOY_MOCK_AGENT}/agent/scope/begin {"name": "<test name>"}
POST {KEPLOY_MOCK_AGENT}/agent/scope/end {"name": "<test name>"}
Authorization: Bearer {KEPLOY_MOCK_AGENT_TOKEN}
At record time this writes a per-test mappings.yaml; at replay time it restricts
the served pool to that test's mocks. No scope calls ⇒ suite-level, which is still
correct. Run the tests one at a time (see below).
The agent's API is authenticated, so a scope call without the header is rejected
with 401 and that test is not scoped. Send the header whenever
KEPLOY_MOCK_AGENT_TOKEN is set, and omit it when it is not, so the same glue
code keeps working against an older Keploy that does not export it — that is what
every example below does.
Watch out for the common shape try: ... except: pass. It hides the 401, and
per-test scoping silently falls back to suite-level while your suite still
passes. Keploy logs the rejection on its own output with the variable to set, so
check there if scoping stops taking effect after an upgrade.
Per-test scoping keeps tests apart only when they run one after another:
go test -p 1 -parallel 1, Playwright --workers=1, jest --runInBand, pytest
without -n. Sending the test process's PID as pid (Node process.pid, Python
os.getpid()) does not keep tests that run at the same time apart. At record, a
call made by any other process (a child the test starts, an app server the tests
share) goes to whichever test was open when it was made; at replay, a process the
test did not start, such as a shared app server, is served the whole recording.
Leave pid out of both calls, as the calls above and every example below do. A
begin without a pid narrows what the whole run is served to the test's mocks,
on every connection. A begin with a pid narrows only the connections that the
test process, or a process it started, opens while a test is open. Pooled
connections the app opened outside a test, such as a database pool it filled at
start-up or between two tests, would be served every test's mocks at replay, and
so would every connection of an app the tests share but did not start, such as
Playwright's webServer or a server started before the suite.
These calls are for runs without KEPLOY_PROVEN=1. Under it Keploy refuses them
(422 upgrade_integration): there a test marks its boundaries through Keploy's
scope protocol v1, which two integrations speak: go-sdk's
keploytest for
go test (go test -p 1), and
@keploy/playwright for
Playwright (--workers=1 --retries=0). Keploy runs one test process at a time
there and refuses a second. There is no integration for other frameworks: their
test process makes the calls by hand. It claims the run
(POST {KEPLOY_MOCK_AGENT}/agent/instance with {"key": "<package folder>"},
the folder relative to the project) before its first test, then sends
each test's begin with "canSkip": true, skips the test when the begin answers
"action": "skip", and sends its end, never with pid, at, dir or suite.
pytest (conftest.py):
import os, json, urllib.request, warnings, pytest
AGENT = os.environ.get("KEPLOY_MOCK_AGENT")
TOKEN = os.environ.get("KEPLOY_MOCK_AGENT_TOKEN")
def _post(path, name):
if not AGENT:
return
body = {"name": name} # run pytest without -n: see above
headers = {"Content-Type": "application/json"}
if TOKEN: # absent on an older Keploy; the call still works there
headers["Authorization"] = "Bearer " + TOKEN
req = urllib.request.Request(AGENT + path, data=json.dumps(body).encode(),
headers=headers, method="POST")
try:
urllib.request.urlopen(req, timeout=3).read()
except Exception as e:
# Surfaced, not swallowed: a 401 here means the test simply is not
# scoped, and silence would make that look like it worked.
warnings.warn(f"keploy: scope call {path} failed: {e}")
@pytest.fixture(autouse=True)
def keploy_scope(request):
_post("/agent/scope/begin", request.node.name)
yield
_post("/agent/scope/end", request.node.name)
go test (TestMain helper):
func scope(path, name string) {
agent := os.Getenv("KEPLOY_MOCK_AGENT")
if agent == "" {
return
}
// Run go test -p 1 -parallel 1: see above.
body, _ := json.Marshal(map[string]any{"name": name})
req, err := http.NewRequest(http.MethodPost, agent+path, bytes.NewReader(body))
if err != nil {
return
}
req.Header.Set("Content-Type", "application/json")
// Absent on an older Keploy, which does not require it.
if tok := os.Getenv("KEPLOY_MOCK_AGENT_TOKEN"); tok != "" {
req.Header.Set("Authorization", "Bearer "+tok)
}
resp, err := http.DefaultClient.Do(req)
if err != nil {
log.Printf("keploy: scope call %s failed: %v", path, err)
return
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
// 401 here means this test is not scoped; do not let that pass quietly.
log.Printf("keploy: scope call %s returned %s", path, resp.Status)
}
}
// In each test: scope("/agent/scope/begin", t.Name()); defer scope("/agent/scope/end", t.Name())
jest / playwright (a beforeEach/afterEach or reporter hook) makes the same
two calls with the test's name. Run jest with --runInBand and Playwright with
--workers=1:
// Playwright: in a fixture or beforeEach/afterEach
const post = async (path, name) => {
const token = process.env.KEPLOY_MOCK_AGENT_TOKEN;
const headers = {"Content-Type": "application/json"};
// Absent on an older Keploy, which does not require it.
if (token) headers.Authorization = `Bearer ${token}`;
try {
const r = await fetch(`${process.env.KEPLOY_MOCK_AGENT}${path}`, {
method: "POST",
headers,
body: JSON.stringify({name}),
});
// A 401 means this test is not scoped — say so rather than swallow it.
if (!r.ok) console.warn(`keploy: scope call ${path} returned ${r.status}`);
} catch (e) {
console.warn(`keploy: scope call ${path} failed: ${e}`);
}
};
test.beforeEach(({}, testInfo) => post("/agent/scope/begin", testInfo.title));
test.afterEach(({}, testInfo) => post("/agent/scope/end", testInfo.title));
Platforms
| Platform | How to run |
|---|---|
| Linux | Native — keploy mock record -c "pytest" (uses eBPF; needs root). |
| Windows | Coming soon. |
| macOS (Apple Silicon) | Native — same command. Userspace interception, so no sudo. Without KEPLOY_PROVEN=1, -c "docker compose run tests" also works. |
Natively on macOS, Keploy 3.9.3 understands HTTP/HTTPS (HTTP/2 included), gRPC,
PostgreSQL, MySQL, MongoDB, Amazon DynamoDB and Aerospike calls; calls to other
services — Redis, Kafka and the like — are captured only as raw bytes and usually
don't replay. If your tests depend on one of those, run them in a container (as
above) or on Linux. The
dependency table lists each one. On
macOS, also
start the test runner itself rather than through a launcher such as npm test or
a wrapper script, and use a Homebrew or uv Python (or a virtualenv built on one),
not Apple's /usr/bin/python3 — see
running natively on macOS.
Keploy doesn't support Intel Macs.
Refresh in CI
Because re-recording overwrites the set in place and the runner's exit code is
propagated, refreshing mocks on a merge to main is a normal CI step:
# bring up the real dependencies, then:
keploy mock record -c "pytest" --name default
keploy sanitize # scrub secrets before committing
git add keploy/ && git commit -m "chore: refresh mocks" || echo "no changes"
By default keploy mock is registry-first: the set is uploaded to Keploy
after record and downloaded before replay automatically (registry use depends on
your plan). Pass --local to keep everything on disk — the --local loop is
also the one that runs without signing in.