Claude Code Week 29: Artifacts Pull Live Data Through MCP Connectors
Claude Code's Week 29 digest (v2.1.207–v2.1.212, July 13–17) headlines with a change to how published Artifacts work: an artifact can now call a user's own MCP connectors every time someone views it, rather than only reflecting a snapshot from the session that built it. A dashboard artifact built to show open pull requests, for example, pulls the live list through the viewer's own GitHub connector on page load — each call runs through the viewing account's own connections, and viewers approve connector access before the first call. The same week added public sharing links for artifacts, editor roles for shared editing on Team and Enterprise plans, and artifacts created directly from Claude Tag sessions.
The second major feature is screen reader mode, which replaces Claude Code's visual terminal interface — boxes, spinners, in-place redraws — with plain, linear text that a screen reader like VoiceOver or NVDA can read top to bottom, including permission prompts. It's available per-session via claude --ax-screen-reader, per-shell via the CLAUDE_AX_SCREEN_READER environment variable, or globally through the axScreenReader setting.
Smaller changes rounded out the week: /fork now copies a conversation into a new background session with its own row in claude agents rather than launching an in-session forked subagent (that behavior moved to /subtask); auto mode no longer requires the CLAUDE_CODE_ENABLE_AUTO_MODE opt-in variable on Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry; and MCP tool calls running longer than two minutes now move to the background automatically to keep the session responsive, tunable via CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS.
Read more — Claude Code Docs
Codex CLI 0.145.0 Stabilizes Multi-Agent V2, Adds Audio Input
OpenAI shipped Codex CLI 0.145.0 on July 21, 2026, stabilizing Multi-agent V2 — configurable sub-agent models, reasoning levels, roles, and concurrency for running several Codex sub-agents on a task at once, previously experimental. The release also introduces audio inputs and tool outputs, including common local audio formats, alongside new streaming realtime V3 conversations for live, voice-driven Codex sessions rather than turn-by-turn text exchanges.
An experimental paginated thread history adds efficient conversation resume, full-text search across past occurrences, persisted thread names, sub-agent support, and memories that persist context across sessions — addressing a common complaint that long-running CLI agent sessions become unwieldy to navigate once a thread grows large. The /import command was also expanded to migrate settings, MCP servers, plugins, sessions, and memories from both Cursor and Claude Code, lowering the switching cost for developers moving between coding agents.
This follows Codex CLI 0.144.5 (July 16), which focused on tightening dangerous rm command detection, and 0.144.6 (July 18), which updated bundled model documentation for OpenAI's GPT-5.6 Sol, Terra, and Luna models.
Read more — OpenAI
GitHub Copilot Vision Reaches General Availability
GitHub Copilot Vision became generally available on July 1, 2026, letting developers attach images and PDFs directly to chat prompts so Copilot can reason about what it sees alongside the surrounding code — a screenshot of a rendering bug, a design mockup, or a PDF spec, for instance. In VS Code, files can be attached by pasting, drag-and-drop, or the context menu. Vision is now on by default across every Copilot plan — Free, Pro, Pro+, Business, and Enterprise — with no admin action required to enable it; Business and Enterprise customers should note GitHub retains image and PDF attachments for approximately 24 hours to provide the service.
Separately, GitHub shipped a new Copilot usage metrics impact dashboard on July 22 for enterprise administrators and organization owners. Where existing usage reporting mostly answered who is active, the new dashboard is built to answer how Copilot is actually being used — aimed at organizations that adopted usage-based billing (which replaced request-based metering across all plans starting June 1, 2026) and now need clearer visibility into what's driving consumption before it shows up as a bill.
Read more — GitHub Changelog
