Claude Code and Cowork Enter Public Beta for Government on FedRAMP High
Anthropic moved Claude Code and Claude Cowork into public beta inside Claude for Government Desktop, built on the same application commercial customers use but delivered through a FedRAMP High authorized environment. Public-sector teams can build and modernize software with Claude Code, while Cowork works directly with files on the desktop to draft memos, review RFPs, handle casework, and build decks.
The controls are the point for agencies: conversation history stored locally on managed devices, inference inside the authorized environment, department-level administration, fixed spending increments, model and seat limits, and tamper-evident audit logs. It deploys through standard agency MDM platforms, and Anthropic remains the contracting and billing party — no separate cloud-provider relationship required.
Claude Adds a “Reflect” Monthly Recap and Focus Settings
Claude rolled out Reflect, a monthly recap at Settings → Reflect that surfaces the topics you spent the most time on, your most active day, your peak hour, and observations about how you work with Claude. Alongside it come focus settings — break reminders and quiet hours — aimed at making heavy Claude use a little more deliberate.
It’s in beta on Free, Pro, and Max plans across web and Claude Desktop, and requires memory to be turned on. Small on its own, but part of a broader move to make Claude aware of usage patterns over time rather than treating every session as a blank slate — and, as some coverage noted, a gentle nudge that keeps you coming back.
MCP’s Largest Revision Since Launch Arrives as a Release Candidate
The Model Context Protocol shipped a release candidate for its biggest revision yet, with the final spec due July 28. The headline is a stateless HTTP core: a remote MCP server that used to need sticky sessions, a shared session store, and deep packet inspection can now run behind a plain round-robin load balancer, route on an Mcp-Method header, and let clients cache tools/list responses via ttlMs and cacheScope — modeled on HTTP Cache-Control.
It also delivers on the 2026 roadmap: MCP Apps (server-rendered UIs), a Tasks extension for long-running work, authorization aligned to OAuth and OpenID Connect, and a formal deprecation policy. Separately, the Enterprise-Managed Authorization extension is now stable, letting organizations gate MCP servers through their identity provider, and servers can advertise themselves via .well-known URLs for easier discovery.
Claude Code Tightens Push/PR Automation and Adds Worktree Safeguards
The latest Claude Code sharpens its agent workflows. /commit-push-pr now auto-allows git push to the repo’s configured push remote (via remote.pushDefault, or the sole remote when there’s only one) in addition to origin, and /cd gained directory-path suggestions. A new /doctor check proposes trimming checked-in CLAUDE.md files by cutting content Claude can derive from the codebase.
The safety-minded fixes matter more as agents run unattended: entering a git worktree outside .claude/worktrees/ now asks for confirmation, an auto-mode rule blocks tampering with session transcript files, a warning fires when your login is about to expire so background sessions don’t die mid-run, and a nasty Windows bug — worktree removal deleting files outside the worktree through an NTFS junction — is fixed.
UST Allies with Anthropic and Commits to Certifying 20,000 Staff on Claude
Digital transformation firm UST announced a strategic alliance with Anthropic focused on helping Global 1000 enterprises become AI-native. The headline commitment: UST plans to certify 20,000 associates worldwide on Claude across engineering, operations, and delivery roles, embedding Claude into its own platforms and services.
It lands as Anthropic’s Claude Partner Network keeps swelling — 40,000+ firms applied, 10,000+ consultants already certified, and $100M committed this year to training and joint go-to-market. Deals like UST are how Claude reaches large enterprises through the systems integrators that actually run their transformations.
The Agent Stack Is Growing Up — Quietly, on Purpose
None of today’s headlines is a flashy new model, and that’s the story. MCP is trading its clever-but-fragile session model for boring, scalable HTTP, plus enterprise auth and a deprecation policy — the unglamorous things a protocol needs before serious companies bet on it. Claude Code is adding confirmations and tamper-blocks exactly where autonomous agents touch your repo. And Anthropic is shipping to government on FedRAMP High with local history and audit logs.
Read together, it’s a maturation beat: the ecosystem is building the guardrails, plumbing, and compliance surface that let agents run unattended in places that can’t tolerate surprises. UST training 20,000 people to deploy Claude is the demand side of the same curve. The exciting phase was watching agents get capable; the durable phase is watching the stack around them get trustworthy. Stateless load balancers and confirmation prompts won’t trend, but they’re what turns a demo into infrastructure.