mcp-handler@2.0.0 now serves the 2026-07-28 MCP spec natively with a backward-compatible Streamable HTTP layer—no Redis or session storage required.
Eliminates infrastructure overhead for MCP server deployments on serverless platforms. Existing Streamable HTTP clients continue working uninterrupted while new clients adopt the stateless protocol.
Replaces @modelcontextprotocol/sdk with @modelcontextprotocol/server and requires Node.js 20+, zod@^4. Breaking change: HTTP+SSE transport removed (/sse, /message return 410 Gone). Worth upgrading now if targeting modern serverless; remain on 1.x if legacy clients depend on SSE.
“mcp-handler@2.0.0 is now available with support for the 2026-07-28 Model Context Protocol specification”
“The stateless 2026-07-28 protocol, served natively, including per-request metadata and server/discover”
“Both protocols with no Redis dependency or session storage”
“Version 2 requires Node.js 20 or later and zod4”
“The deprecated HTTP+SSE transport has been removed. Requests to /sse and /message return 410 Gone”
Get issues like this in your inbox — free, every weekday.
Quick Signals
GPT-5.6 Luna drops 80%, Sol 2.5x faster
AI Gateway passes through upstream price cuts and speed gains with zero markup; existing code gets new rates automatically.
Token economics shift materially for Luna workloads (80% reduction enables new use cases), and Sol's fast mode speed bump directly improves latency-sensitive inference without code changes. Model IDs unchanged means you get the upgrade on next request.
Replace nothing—this is a pure improvement to existing AI Gateway routes. No code changes required. Act now if you're on Luna (lock in 80% savings) or if latency is your bottleneck (Sol fast mode now 2.5x faster). Worth immediate cost re-audit for Luna-heavy deployments.
“AI Gateway adds no markup on token pricing, so these changes reach you at the upstream rate”
“GPT-5.6 Luna: 80% price reduction”
“GPT-5.6 Sol: fast mode now 2.5x faster (up from 1.5x)”
“Model IDs are unchanged, so existing requests get the new rates and speed with no code change”
pricinglatencygpt-5.6ai-gatewaycost-optimization
Inkling Small ships on AI Gateway with reasoning
Quarter-size model matching Inkling performance, controllable thinking effort, native multimodal reasoning—set model to `thinkingmachines/inkling-small` in AI SDK.
Reduces inference cost and latency for agentic coding and tool-use workflows while maintaining reasoning quality. Zero Data Retention option available per-request, enabling compliance-first deployments without vendor lock-in.
Drops into existing AI SDK calls as a model swap. Requires no setup beyond selecting the model identifier and optionally enabling ZDR flag. Worth trying now for cost-sensitive coding agents and multimodal document workflows; benchmark against your current model first.
Data Point
Evidence ledgers route unsupported AI claims to authors
Pair each claim with traced evidence packets and auto-route contradictions back to the author—0.676 relation accuracy on blind benchmark—turning heterogeneous sources into auditable traceability for AI-assisted writing workflows.
Developers building AI writing assistants need verifiable claim-to-evidence chains; this workflow catches hallucinations and mixed evidence before publication, reducing post-hoc fact-checking load and establishing clear responsibility boundaries.
Replaces manual spot-checking with systematic evidence routing. Requires labeled evidence packets, relation classifiers trained on AVeriTeC/CLIMATE-FEVER/SciFact patterns, and routing logic to surface problematic claims. Worth prototyping now if you control the claim generation pipeline; production-ready for constrained domains.
“evidence-ledger adjudication: a claim-evidence traceability workflow that pairs each claim with an evidence packet, assigns a support relation, and routes unsupported, contradicted, or mixed-evidence claims back to the author”
Rails Active Storage vips RCE requires immediate patching
Default Rails 7.x/8.x with Active Storage + vips processor exposes arbitrary file read and RCE if your app accepts untrusted image uploads; patch to 7.2.3.2, 8.0.5.1, or 8.1.3.1 immediately, and verify libvips ≥ 8.13.
Active Storage ships enabled by default in production Rails apps. This vulnerability chains image upload handling to code execution without requiring authentication in some scenarios, making it a runtime blocker for any Rails service processing user-uploaded images.
Upgrade Rails to patched versions (7.2.3.2+, 8.0.5.1+, 8.1.3.1+). Separately verify libvips ≥ 8.13; earlier versions cannot disable unsafe operations even after Rails patch. If upgrade blocked, set VIPS_BLOCK_UNTRUSTED=1 or call Vips.block_untrusted(true) as temporary mitigation. WAF is insufficient. Do not delay—PoC reconstruction is likely once diffs go public.
“Arbitrary file read and remote code execution chain affecting default Rails applications that use Active Storage with the vips image processor”
“move to Rails 7.2.3.2, 8.0.5.1, or 8.1.3.1 (or later)”
“the fix depends on libvips itself being version 8.13 or later, earlier libvips builds can't disable the unsafe operations involved at all”
“Rails 7.0.0 through 7.2.3.1 and 8.0.0 through 8.1.3 don't have that requirement”
“assumes attackers with AI tooling can reconstruct one quickly once patches ship and diffs go public”
rails-securityactive-storagercevipspatch-urgent
Vercel Sandbox runs isolated agents in single environment
Use createUser() and createGroup() to run multi-agent systems with filesystem isolation and optional shared workspaces in @vercel/sandbox.
Eliminates the operational overhead of spinning up separate containers per agent while maintaining security boundaries. Developers building multi-agent workflows can now coordinate work through shared directories without sacrificing process isolation.
Replaces single-user sandbox patterns and replaces running agents in separate container instances. Requires Vercel Sandbox SDK v1.x with multi-user support; ready to adopt now for new multi-agent projects. Existing single-user code remains compatible.
“Each agent runs as its own user with a private home directory”
“users can't read, write, or list each other's files”
“A group opens a shared workspace when they need to collaborate”
“Call createUser for each agent; its commands and file operations run as that user”
Upgrade to Nuxt 4.5.1 or 3.21.10 immediately—high-severity RCE in server islands and authorization bypass in route rules require patched versions, Vercel WAF mitigations are not a replacement.
Server-side RCE and cross-user cache disclosure directly compromise production apps; relying on platform-level protections without upgrading leaves component instantiation, caching, and DoS vectors exposed. Non-negotiable dependency bump for any Nuxt deployment handling authenticated data.
Replace current Nuxt versions with patched releases via `npx nuxt upgrade --dedupe`—requires lockfile refresh and DevTools bump to 3.3.1. Critical for production; no workaround exists. Also required: purge upstream caches if serving authenticated payloads pre-patch. Start now.
“eight security advisories, including a high-severity server-side remote code execution vulnerability”
“Nuxt 4: `4.5.1` or later, Nuxt 3: `3.21.10` or later, Nuxt DevTools: `3.3.1` or later”
“Vercel received advance notice of the server-side remote code execution vulnerability, [GHSA-9473-5f9j-94wq], and deployed platform-wide WAF mitigations before public disclosure”
“do not rely on them for full protection. Upgrading to a patched version is still required”
“The newly disclosed authorization bypass is a regression in that earlier fix”