Daily AI Builder Brief - April 24, 2026

Executive Summary

Claude Code 2.1.119 ships a 40+ item fix release with real MCP/OAuth repairs and better headless runs.
--print mode now respects an agent's tools: and disallowedTools: frontmatter, and --agent <name> honors that agent's permissionMode.
MCP HTTP connections no longer fail with bogus OAuth error responses, and ${ENV_VAR} placeholders in headers finally expand for HTTP, SSE, and WebSocket servers.
--from-pr accepts GitLab MR, Bitbucket PR, and GitHub Enterprise PR URLs; owner/repo#N shorthand uses the git remote's host.
Hooks: PostToolUse and PostToolUseFailure now include duration_ms; OpenTelemetry tool_result and tool_decision events carry tool_use_id.
Claude Managed Agents gains built-in memory in public beta under the managed-agents-2026-04-01 header, so agents retain state across sessions.
Anthropic and NEC announce a strategic collaboration to build one of Japan's largest AI-native engineering orgs, with Claude Code rolling out to roughly 30,000 NEC Group staff.
MCP spec is still the November 2025 release. No new protocol cut this week.

Claude Code

v2.1.119 (April 23)

Headless and agent paths get the bulk of the attention. --print enforces agent tools/disallowedTools, --agent uses the agent's permissionMode, and PowerShell tool commands can be auto-approved. blockedMarketplaces actually enforces hostPattern and pathPattern. MCP fixes: OAuth HTTP error parsing, ${ENV_VAR} in headers, and OAuth client secret sent during token exchange. Subagent and SDK MCP servers reconnect in parallel. /config persists to ~/.claude/settings.json with project/local/policy precedence. New prUrlTemplate setting and CLAUDE_CODE_HIDE_CWD env var. Changelog

Anthropic

Claude Managed Agents: built-in memory (April 23)

Persistent memory across sessions, available today in public beta under the managed-agents-2026-04-01 beta header. Useful when an agent needs to remember user preferences, prior tool outputs, or task state between runs without you rolling your own store. Announcement

Anthropic + NEC strategic collaboration (April 23/24)

NEC becomes Anthropic's first Japan-based global partner. Roughly 30,000 NEC Group staff get Claude, with Claude Code at the center of what NEC is calling one of Japan's largest AI-native engineering organizations. Joint go-to-market targets finance, manufacturing, and local government, with Claude Cowork as the desktop agent surface. Anthropic post · NEC release

MCP Ecosystem

No new spec or official server

Spec is still 2025-11-25. 2026 roadmap still centers on streamable HTTP scaling, Tasks primitive lifecycle, and enterprise auth, but nothing cut in the last 48 hours. MCP Blog

Broader AI

Nothing today.

What This Means For Builders

If you script Claude Code with --print or --agent in CI, re-audit your agent frontmatter. Tool and permission scoping now actually applies, so previously over-permissive agents may start refusing work.
The MCP OAuth and ${ENV_VAR} header fixes unblock a real class of "works locally, fails in prod" bugs. Retry any previously broken remote MCP integrations.
If you were building your own agent memory layer on top of the API, evaluate Managed Agents memory before investing another sprint.