GET /v1/experiments

Fetch experiment variant assignments for a user. All six SDKs (Swift, Android, React Native, JavaScript, Flutter, Capacitor) call this for you — use it directly for server-side assignment or platforms without an SDK.

Request

curl "https://ingest.mostlygoodmetrics.com/v1/experiments?user_id=user_123&anonymous_id=550e8400-..." \
  -H "Authorization: Bearer mgm_proj_your_api_key"

Query parameters:

Parameter	Required	Description
user_id	No	The user to assign variants for.
anonymous_id	No	The client's stored anonymous ID. Can be sent alone (pre-login) or together with user_id.

When both are sent, the server links the two IDs (idempotently) before assigning, so a user who was bucketed while anonymous keeps the same variant after identify() instead of being re-bucketed. Always send the stored anonymous ID alongside user_id on the refetch after login.

Responses

Always 200 OK.

With user_id and/or anonymous_id — assigned variants per active experiment:

{
  "assigned_variants": {
    "button-color": "a",
    "pricing-test": "control"
  }
}

Assignments are sticky: the same ID always gets the same variant for a running experiment.

With no parameters (legacy) — the list of active experiments and their variants:

{
  "experiments": [
    { "id": "button-color", "variants": ["a", "b"] },
    { "id": "pricing-test", "variants": ["control", "low", "high"] }
  ]
}

Errors: 401 {"error": "Invalid or missing API key"}.

GET /v1/experiments/configs

Fetch the configurations of running experiments — ID, name, and variant list — with no user parameters. This powers local experiment enrollment: SDKs in local mode fetch the configs and bucket the user on device, so no user identifier is ever sent to the server for assignment.

curl "https://ingest.mostlygoodmetrics.com/v1/experiments/configs" \
  -H "Authorization: Bearer mgm_proj_your_api_key"

Authentication is the same project API key; the endpoint accepts no query parameters and ignores any sent.

Response — 200 OK with only running experiments:

{
  "experiments": [
    {
      "id": "7b1e8a90-4c2d-4f6a-9e3b-2a1d5c8f0e71",
      "name": "button-color",
      "variants": ["control", "treatment"]
    }
  ]
}

id is the experiment's UUID — it is what on-device bucketing hashes. To replicate the SDKs' deterministic assignment: take the first 8 bytes of SHA-256("<experiment_uuid>:<user_id>") as a big-endian unsigned 64-bit integer and pick variants[bucket % variants.length]. Same errors as above (401 for a bad key).

Reporting exposure

For results to count, the server needs to know when a user actually experienced a variant. Send an exposure event via POST /v1/events the first time your code acts on an assignment (once per experiment + variant per user):

{
  "name": "$experiment_exposure",
  "user_id": "user_123",
  "properties": {
    "$experiment_name": "button-color",
    "$variant": "a"
  }
}

$experiment_name is the experiment name as returned by this endpoint (the snake_cased form is also accepted).

Optionally, also attach the assignment to every event as a super property, $experiment_{snake_case(name)}: "variant" (e.g. $experiment_button_color: "a"). The SDKs do both automatically.

See Experiments for how enrollment and conversions are counted.