Claude Code Daily Briefing - 2026-04-29

Release Summary

VersionDateKey Changes
v2.1.1224/28ANTHROPIC_BEDROCK_SERVICE_TIER, PR URL pasting in /resume, image resize fix
v2.1.1214/28MCP alwaysLoad option, plugin prune, ~2GB memory leak fix, hook output replacement

After the v2.1.120 recall, v2.1.121 and v2.1.122 shipped back-to-back on the same day — ending the 5-day regression period and returning to a stable release cadence.

Full Release Notes

New Features & Practical Usage

v2.1.121 — MCP alwaysLoad, ~2GB Memory Leak Fix, Hook Output Replacement (4/28)

Following the v2.1.120 recall and stabilization work, v2.1.121 brings significant features and performance fixes.

MCP alwaysLoad option: Set alwaysLoad: true in your MCP server config to make all tools from that server skip tool-search deferral and remain always available. This solves the problem of frequently-used MCP tools being missed by lazy loading.

~2GB memory leak fix: Fixed unbounded memory growth when processing many images in a session, and a separate leak where /usage could consume up to ~2GB on machines with large transcript histories. Long-session users will notice meaningful performance improvements.

PostToolUse hook output replacement: hookSpecificOutput.updatedToolOutput now works for all tools, not just MCP tools. You can now replace output from Bash, Read, Grep, and other built-in tools via hooks.

// settings.json — Always load an MCP server
{
  "mcpServers": {
    "my-critical-server": {
      "command": "npx",
      "args": ["-y", "@my/mcp-server"],
      "alwaysLoad": true
    }
  }
}

v2.1.122 — Bedrock Service Tier, /resume PR URL Support (4/28)

Released shortly after v2.1.121, v2.1.122 adds features for AWS Bedrock users.

ANTHROPIC_BEDROCK_SERVICE_TIER: Choose between default, flex, or priority tiers on Bedrock, balancing cost against latency.

PR URL pasting in /resume: Paste a PR URL into /resume to automatically find and continue the session that created it. Supports GitHub, GitLab, and Bitbucket.

GitHub v2.1.121 | GitHub v2.1.122

Developer Workflow Tips

Filter Sensitive Output with PostToolUse Hooks

The expanded updatedToolOutput capability in v2.1.121 enables practical security patterns. You can automatically mask API keys, tokens, and internal URLs from Bash tool output, or summarize excessively long outputs to save context tokens. Since this now works for all built-in tools (not just MCP), organizations can enforce security policies at the hook level across the entire tool surface.

GitHub v2.1.121

Self-Updating Screenshots in Documentation

A system that embeds SCREENSHOT directives in Markdown comments, then captures screenshots via headless Chrome during the documentation build. It supports element, full_page, and viewport modes with options like click, wait, and crop to control UI state.

This solves the common problem of outdated screenshots after Claude Code modifies your UI. A single build command handles both screenshot capture and documentation generation, making it easy to integrate into CI/CD pipelines.

Original Post

Security & Limitations

Claude.ai Major Outage — 12,000+ Users Affected, ~78 Minutes Down (4/28)

On April 28 (UTC 17:34–18:52), Claude.ai, Claude Code, and the Anthropic API simultaneously went down. Over 12,000 outage reports were logged on Downdetector, and Anthropic’s status page classified it as a “Major Outage.”

The primary symptoms were authentication errors affecting login paths, including Claude Code login. Claude Design and other connected services were also impacted. Service was restored after approximately 78 minutes — but notably, a similar major outage occurred just 8 days prior (April 20), raising concerns about infrastructure reliability.

For Claude Code users, actively using --resume and periodically saving work state during critical sessions is recommended as a mitigation strategy.

Tom’s Guide | GV Wire | Claude Status

Ecosystem & Plugins

GitHub Copilot Shifts to Usage-Based Billing — Effective June 1 (4/28)

All GitHub Copilot plans will transition to monthly AI Credits-based usage billing starting June 1, 2026. Instead of per-request pricing, billing will be based on input/output/cached token consumption. Base prices remain the same (Pro $10/mo, Pro+ $39/mo), but there’s no fallback when credits run out — features simply stop working. Code completions remain credit-free.

This could change the competitive pricing dynamics against Claude Code’s Max plan ($200/mo), as actual costs will now depend heavily on usage patterns.

GitHub Blog | GeekNews

Microsoft and OpenAI End Exclusivity and Revenue Sharing (4/27)

Microsoft and OpenAI have dropped their exclusive distribution and revenue-sharing agreements. OpenAI can now contract directly with competing clouds like AWS, while Microsoft stops paying revenue share on Azure resales.

This diversifies OpenAI’s distribution channels and adds complexity to the infrastructure competition between OpenAI (now multi-cloud) and Anthropic (AWS + Google Cloud).

Bloomberg | GeekNews

TrafficMonitor — Claude/Codex Usage Limits in Your Windows Taskbar

A Windows TrafficMonitor plugin that displays Claude and Codex usage limits in real-time on your taskbar. Monitors 5-hour and 7-day usage windows, with all data processed locally. Useful for Max plan users who want to avoid hitting usage limits unexpectedly.

GeekNews | GitHub

Community News

Minor Changes

Interesting Projects & Tools

By subscribing, you agree to our Privacy Policy.