v4 is available alongside v3. Existing integrations on v3 keep working; migrate resource by resource when convenient. Each endpoint in the guides below is tagged New or Updated with a link back to its v3 counterpart.
Designed for agents and MCP
v4 was built with AI agents and the Model Context Protocol in mind. Most v4 resources are also exposed through the official Qonversion MCP server, so they are callable directly from Claude, ChatGPT, Cursor, and any other MCP-compatible client — without a custom wrapper. Concretely, v4 follows a set of rules that make it safe and predictable for an agent to use:- Stable, verb-based operation IDs (e.g.,
v4CreateScreen,v4ListExperiments) — most MCP tools wrap a single v4 operation. - Consistent resource-oriented URLs —
/v4/{resource}for collections,/v4/{resource}/{id}for items, sub-resources nested only one level deep. - Pagination varies by endpoint. Products, entitlement definitions, offerings, segments, remote configurations, screens, automations, experiments, customers, and user purchases use
limitwith astarting_aftercursor — pass thenext_cursorvalue from the previous page unchanged. Events and scheduled reports uselimitwithoffset. Some lists, including a user’s entitlements and integrations, return the complete set without pagination. Read the endpoint before writing a generic paginator. - Idempotent mutations where feasible (PUT, PATCH, and DELETE are safe to retry).
- One token type over HTTP — v4 public endpoints authenticate with the project Secret Key; the MCP server uses OAuth 2.1 with per-tool scopes instead, so AI clients never handle the Secret Key directly. One operation needs a signed-in dashboard user and is therefore available through the MCP server only: switching off a store integration (
enabled: falseonPOST /v4/project-settings/stores/{platform}). With a Secret Key alone, it returns403control_forbidden. - Machine-readable errors — every error includes
type,code, and field-level details so an agent can self-correct without parsing prose.
Base URL
All v4 API requests should be made to:API Sections
Request Format
All requests must include:Authorizationheader with a Bearer token (see Authentication)Content-Type: application/jsonforPOST,PUT,PATCH,DELETErequests