Changelog

First release of the Python SDK and the MCP server. Backend hardening + dashboard precision fixes from the live audit.

• Python SDK published to PyPI as weckr-sdk@0.1.1 (companion to @weckr/sdk@0.1.3)
• @weckr/mcp@0.1.0 — Model Context Protocol server so Claude, Cursor & co. can query your cost data
• Spending caps (block or downgrade) per plan with same-provider model swaps
• Pricing recommendations — suggests what to charge based on real usage data
• Model recommendations — detects expensive models on simple-output tasks
• WeckrConfigError (new) — 401/403 from /check now fails CLOSED so misconfig surfaces immediately
• wk.flush() / wk.flush() — wait for in-flight log POSTs before short-lived processes exit
• Dated model variants (gpt-4o-2024-08-06, claude-3-5-sonnet-latest, …) now price correctly via longest-prefix match
• Server: dropped per-row margin column, derive margin = sum(revenue) − sum(cost) at full precision in read RPCs
• Security: DEV_API_KEY now hard-gated by NODE_ENV; per-key rate limits + project-id verification on dashboard reads

Initial release.

• @weckr/sdk published to npm
• OpenAI, Anthropic, and Gemini support out of the box
• Per-user margin tracking via plan_revenue − cost
• Dashboard at app.useweckr.com — Overview, Users, Features, Recommendations