RepoJournal
Vercel

@vercel

Next.js + the Vercel platform - frontend infrastructure for the web

Keep up with Vercel in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

One email a day. Unsubscribe in one click.

Pick a date

Topics: JavaScript Full archive →

The Wire · Showcase

MCP OAuth hardening lands, Fireworks structured outputs fixed

By RepoJournal · Filed · About Vercel

AI SDK patches a security hole in OAuth client registration while unlocking structured outputs for Fireworks models.

Vercel's AI SDK shipped a breaking MCP update hardening OAuth client registration against the latest protocol requirements, including RFC 9207 `iss` validation and mandatory `application_type` [1]. This pairs with a Fireworks fix that finally enables structured outputs for chat models, ending silent degradation of JSON schema requests [2]. And if you've wrestled with LangGraph tool-call ID collisions, the SDK now preserves tool lifecycles when provider IDs repeat across steps [3]. Over in the workflow engine, a 100-line core diff pins draw order to event-log order, killing the residual `CORRUPTED_EVENT_LOG` class that plagued slot-mode runs [4]. For Terraform users, the provider now imports webhooks without forcing secret rotation and keeps firewall config IDs intact [5][6]. The release notes for the firewall fix say the provider can "mistake the edit for a removal plus insertion," but that's fixed too [7]. Rounding out the stack, Eve's approval flow stops duplicating pending approvals and restores Slack approval settlement feedback [8][9], while vercel-py lets you attach metadata to hooks for richer workflow context [10]. Across 22 stories and 51 commits, this is a day of correctness and security.

One email a day. Unsubscribe in one click.

Action items

References

  1. [1] feat(mcp): harden oauth client registration according to latest protocol (#19043) vercel/ai
  2. [2] fix(fireworks): enable structured outputs for chat models (#19025) vercel/ai
  3. [3] fix: preserve LangGraph tool lifecycles when provider tool call IDs repeat across steps (#18837) vercel/ai
  4. [4] [core] Pin draw order to event-log order: retire delivery barriers through one ordered dispenser ↗ vercel/workflow
  5. [5] Add import support for Vercel webhooks ↗ vercel/terraform-provider-vercel
  6. [6] Keep project IDs in firewall config state ↗ vercel/terraform-provider-vercel
  7. [7] Update edited firewall rules in place ↗ vercel/terraform-provider-vercel
  8. [8] fix(eve): prevent duplicate pending approvals ↗ vercel/eve
  9. [9] fix(eve): wait for Slack approval settlement ↗ vercel/eve
  10. [10] workflow: attach metadata to a hook ↗ vercel/vercel-py

Quick answers

What shipped in Vercel on August 19, 2026?
AI SDK patches a security hole in OAuth client registration while unlocking structured outputs for Fireworks models. In total, 51 commits, 55 pull requests, and 12 releases landed.
Who contributed to Vercel on August 19, 2026?
15 developers shipped this update, including Aayush Kapoor, ai-sdk-factory[bot], Brice Lechatellier, Kevin Dawkins, Mohamed Rebai, mitul-s, alangenfeld, and VaguelySerious, and 7 more.
What were the notable Vercel updates?
feat(mcp): harden oauth client registration according to latest protocol (#19043), fix(fireworks): enable structured outputs for chat models (#19025), and fix: preserve LangGraph tool lifecycles when provider tool call IDs repeat across steps (#18837).

More from @vercel

Daily updates, in your inbox

Follow Vercel

Keep up with Vercel in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

One email a day. Unsubscribe in one click. Read a past issue →

Elsewhere on the wire

Want every project, not just this one?