MCP's Breaking Change, a Hallucinating WebFetch, and MiniMax H3 Goes Low-VRAM

Today's findings

1. #1 Free checker for MCP's breaking 2026-07-28 protocol revisiontool

   A deterministic, rule-based checker (web demo + Claude Skill) grades your MCP server against the 2026-07-28 spec — stateless transport, no more Mcp-Session-Id, OAuth 2.1 — instead of an LLM guessing.

   MCP · 2026-07-28 revision

   Seven breaking rules your MCP server is now graded on

   7

   breaking rules in the new spec

   Biggest breaking change MCP has shipped

   Stateless

   transport now required

   Dropped

   Mcp-Session-Id header

   OAuth 2.1

   new auth baseline

   A free rule-based checker (web demo + Claude Skill) returns a spec-referenced verdict — no LLM guessing.

   Why it matters: This is the biggest breaking change MCP has had, and public servers are quietly broken right now; if your team exposes any MCP tools, migrating is a refactor, not a version bump.

   How to apply: Paste your MCP endpoint into the web demo or install the Claude Skill to get a spec-referenced report of exactly which of the 7 breaking rules you violate, before a client silently fails on you.

   mcpclaude-codeprotocol

   Read more: I built a free Claude Skill + checker for MCP's 2026-07-28 breaking change - and what I learned making it

2. #2 Claude Code's WebFetch quietly delegates to a smaller, hallucination-prone modeltip

   WebFetch in Claude Code summarizes pages via a cheaper subagent model before Opus ever reads them, and that model can invent stats, quotes, and citations that don't exist.

   Claude Code · WebFetch

   A cheaper model reads the page — Opus only ever sees its summary

   1

   WebFetch

   you pass a URL

   2

   Subagent

   reads the page

   3

   Summary

   may invent facts

   Lossy hop: stats, quotes and citations can be fabricated here

   4

   Opus

   never sees source

   Treat WebFetch output as a summary, not a read — spot-check suspiciously precise numbers.

   Why it matters: Anyone using Claude Code for research, competitive analysis, or sourcing claims in docs or PRs risks shipping fabricated facts that read as confidently cited.

   How to apply: Treat WebFetch output as a lossy summary, not a verified read — spot-check tool outputs for suspiciously specific numbers, and directly ask Claude whether it actually read the source when precision matters.

   claude-codeagentshallucination

   Read more: PSA: Be careful letting Claude use WebFetch for research 😵‍💫

3. #3 Offline code index resolves imports for coding agents without a language serverrepo

   An MIT-licensed Rust tool builds a call-graph-aware code index for AI coding agents with no rust-analyzer, no LSP, and no build step — at the cost of LSP-level precision on generics/macros/DI.

   Code intelligence for agents

   Drop the language server, keep the call graph

   LSP / rust-analyzer

   • Needs a working build
   • Exact on generics + macros
   • Follows DI wiring
   • Silent when the repo won't compile

   vs

   Offline code index

   • No LSP, no build step
   • Indexes broken code
   • Call graph queried over MCP
   • Fuzzier on generics, macros, DI

   Trades LSP-grade precision for a persistent 'what calls this' answer instead of a dozen file reads

   MIT-licensed Rust indexer wired into coding agents via MCP

   Why it matters: Coding agents burn huge context re-reading files to answer 'what breaks if I change this signature'; a persistent index over MCP turns that into a structured query and still works when the repo doesn't currently compile.

   How to apply: Wire it into your coding-agent setup over MCP for large monorepos where 'what calls this' questions currently mean the agent reading a dozen files first.

   coding-agentsmcprepo

   Read more: A local code index for coding agents that resolves imports without a language server (Rust, MIT, runs offline)

4. #4 llama.cpp multi-GPU tensor-split crashes fixed by setting -ub 384tip

   A GPU-memory-corruption bug in llama.cpp's tensor-split path (hit on multi-card AMD gfx1030 setups over both ROCm and Vulkan) is worked around by setting the ubatch size to 384 instead of the 512 default.

   llama.cpp · multi-GPU workaround

   Dropping ubatch from 512 to 384 stops tensor-split VRAM corruption

   -ub = 384 the 512 default triggers the memory bug

   -b = 384 × N GPUs batch must stay a multiple of ubatch

   --tensor-split = enabled the path where the crashes appear

   Reported on multi-card AMD gfx1030 rigs under both ROCm and Vulkan.

   Why it matters: If you're pooling multiple older or mid-range GPUs for local inference, this is the difference between tensor-split working at all and constant crashes — a one-flag fix that's easy to miss.

   How to apply: Add `-ub 384` and set `-b` to a multiple of 384 times your GPU count when running llama.cpp with `--tensor-split` across multiple cards.

   llama.cpplocal-llm

   Read more: PSA for anyone with multiple V620's or other gfx1030 cards having problems making llama.cpp tensor split work -- set "-ub 384" and -b to a multiple of that depending on number of GPUs

5. #5 Qwen3.6 35B-A3B MoE beats 27B dense on speed with a smaller quality gap than expectedtechnique

   Local coding tests found the 35B-A3B MoE variant ~4x faster (~116 vs ~30 tok/s) than the 27B dense model, with dense's edge showing up mainly on implicit invariants and edge cases, not basic correctness.

   Local coding models

   MoE runs ~4x faster; dense only wins on the subtle stuff

   vs

   Qwen3.6 35B-A3B (MoE)

   27B dense

   Throughput

   ~116 tok/s

   ~30 tok/s

   Speed multiple

   ~4x

   1x baseline

   Basic correctness

   Holds up

   Holds up

   Implicit invariants

   Misses more

   Catches more

   Edge cases

   Weaker

   Stronger

   Routine edits

   Best default

   Slow for the gain

   Qwen3.6 35B-A3B (MoE) wins the row 27B dense wins the row

   Default agent loops to MoE; route edge-case-heavy refactors to dense.

   Why it matters: For teams choosing a local coding model, this suggests MoE is the better default for throughput, reserving dense models for the trickiest review passes rather than everyday generation.

   How to apply: Default agent loops to the MoE quant for routine edits and bug fixes on local hardware, and route only edge-case-heavy refactors to a dense model or frontier fallback.

   local-llmcoding-agents

   Read more: Qwen 35B-A3B MoE vs 27B dense in local coding tests: ~4× faster, much smaller quality gap than I expected

6. #6 llama.cpp adds in-progress support for Longcat-Flashrepo

   A pending PR brings the Longcat-Flash MoE model to llama.cpp and is actively seeking community testing before merge.

   llama.cpp · work in progress

   Longcat-Flash MoE support is up as a PR — testers wanted before it merges

   Longcat-Flash

   model PR open · unmerged

   Open-weight MoE → GGUFPending merge

   PR branchGitHub review threadGGUF quants (after merge)

   rungh pr checkout … && build llama.cpp

   Run it against your usual local eval suite and post results on the thread.

   Why it matters: Testing now means faster GGUF availability and fewer surprises once it lands — worth a try if you're evaluating new open-weight MoE options for local deployment.

   How to apply: Pull the PR branch, run it against your usual local eval suite, and report results on the GitHub thread while it's still shaping what ships.

   llama.cpplocal-llmgguf

   Read more: model: support Longcat-Flash (need testing) by ngxson · Pull Request #19182 · ggml-org/llama.cpp

7. #7 4-bit (w4a8) quantization brings MiniMax H3 video generation to 8GB VRAMtool

   kijai released a w4a8 (int4 weights, int8 activations) quantization of the open-weights MiniMax H3 video model that runs at int8-or-better speed on 8GB-and-under GPUs.

   Why it matters: MiniMax H3 has been the most-discussed open video model this week but was previously out of reach for lower-VRAM local setups; this quant closes that gap without the usual low-bit speed penalty.

   How to apply: Swap the w4a8 checkpoint into existing ComfyUI MiniMax H3 workflows if you're experimenting with local video generation on consumer GPUs.

   quantizationvideo-generation

   Read more: For 8GB (& below) Vram Rockers, kijay released a 4 bit version of minimax

8. #8 Fix for MiniMax H3 black-frame bug on pre-Ampere GPUs: ~11x speeduptechnique

   Forcing fp16 on MiniMax H3 caused black frames because three DiT submodules overflow fp16's range; keeping only those in fp32 fixes it with no quality loss and ~11x the speed of the fp32 fallback.

   MiniMax H3 · Volta-class GPUs

   Keeping three DiT submodules in fp32 fixes the black frames at ~11x the speed of full fp32

   Full fp32 fallback

   • 330-370 s per step
   • Forced fp16 → black frames
   • Silent fallback: V100, P100, Titan V

   → ~11x faster · no quality loss

   fp32 only where it overflows

   • Text-conditioning projection
   • Attention-sink rows
   • Flagged block outputs

   Everything outside those three submodules runs fp16; patch targets ComfyUI on cards without bf16.

   Why it matters: Cards without bf16 support silently fall back to full fp32 in ComfyUI and crawl at 330-370s/step — this is a targeted, verified fix rather than a full-precision workaround for anyone stuck on Volta-class hardware.

   How to apply: Apply the mixed-precision patch (fp32 only on the text-conditioning projection, attention-sink rows, and flagged block outputs; fp16 elsewhere) if running MiniMax H3 on a V100, P100, or Titan V.

   video-generationquantization

   Read more: MiniMax H3 at native fp16 on V100 ~11× faster, fixes the black frames

9. #9 Mistral open-sources Shieldstral 1.0 3B, a policy-adaptive safety classifiertool

   Shieldstral 1.0 is an open-weights 3B multimodal safety/moderation classifier that Mistral claims matches models 7x its size, taking a text policy description at inference time rather than requiring retraining.

   Why it matters: A small, open, policy-adaptive moderation model you can self-host is directly useful for teams needing content filtering on user uploads or agent outputs without a third-party moderation API.

   How to apply: Evaluate it as a local pre-filter in front of user-generated content or agent tool outputs, feeding it your own policy text, then benchmark against whatever moderation API you currently pay for.

   safetylocal-llm

   Read more: Mistral AI Releases Shieldstral 1.0 3B: An Open-Weights Policy-Adaptive Multimodal Safety Classifier Matching Models 7× Its Size

10. #10 Claude Code 2.1.224 lets sessions message each other — and it's already raising prompt-injection alarmstool

    The new release adds agent-to-agent messaging so separate Claude Code sessions can hand off context directly, but critics immediately flagged it as a transport layer for prompt-injection to spread between agents.

    Claude Code 2.1.224

    Session-to-session messaging is a genuine handoff channel — and a genuine injection surface

    bounded capability

    Agent → agent messaging

    scope Allowlist pairs

    limit Handoffs only

    limit No untrusted parsers

    monitor Log every message

    A compromised agent can now instruct another agent by design.

    Why it matters: If you run multi-terminal or multi-agent Claude Code setups, this changes your workflow — but any inter-agent channel is an injection surface, since a compromised agent can now message instructions to another by design.

    How to apply: Use it for legitimate handoffs (e.g. a backend agent notifying a frontend agent of an API change), but restrict which agents can message which, and don't let an agent parsing untrusted external content also hold messaging permissions.

    claude-codeagentsprompt-injection

    Read more: Claude code : your sessions can now message each other. · Claude Code 2.1.224 - inter-agent messaging: the transport layer for AI worms

11. #11 Custom sigma schedule reduces warping and boosts detail in MiniMax H3 generationstip

    A community-tuned 24-step sigma schedule for MiniMax H3 cuts warping and adds detail versus the default scheduler, with deis/heun samplers giving a further bump at the cost of generation time.

    Why it matters: Free scheduler tweaks like this compound across every video generation you run locally, with no retraining or extra VRAM required.

    How to apply: Drop the published sigma values into your MiniMax H3 ComfyUI sampler node in place of the default schedule, and switch to deis or heun when you want maximum quality and can tolerate slower steps.

    video-generationsampling

    Read more: Some custom sigmas for better detail and less warping