Changelog

Notable changes to the Pushrail API and SDKs land here.

2026-06-04

Account-wide MCP API keys, a new key type (prefix pmk_) for connecting AI agents, the CLI, and the MCP server. Minted at MCP API keys in the dashboard with a read or manage scope; not tied to an environment. Distinct from the environment-scoped API keys used to send events. See MCP API keys.
Per-call environment for agents, MCP tools that touch environment-scoped resources now take an environment argument per call — required on writes (create_destination, send_test_event), optional on reads (omit to list across all environments). Connecting an agent no longer asks you to pick an environment. See Specifying an environment per call.
Production guardrail, writes targeting Production return a warning and the MCP server instructs agents to confirm before any production write.

CLI, the pushrail command-line tool: manage destinations and customers, send and replay events, stream live deliveries to a local endpoint (pushrail listen), inspect metrics/logs/audit/DLQ, run CI checks and smoke tests, and mint portal links. See the CLI quickstart.
MCP server, a hosted Model Context Protocol endpoint so AI agents can operate your workspace with scoped, safety-gated tools.
API key scopes, keys now carry an explicit scope (ingest, manage, or read), enforced per endpoint. See Authentication.
GET /auth/context, introspect the calling credential's workspace, environment, scopes, and permissions.