--- title: Claude Code gets auto mode, Figma ships native MCP write... description: Anthropic shipped auto mode for Claude Code, a research preview for Team plan users that uses a classifier to automatically allow safe actions and block... canonical: https://daily.dev/posts/claude-code-gets-auto-mode-figma-ships-native-mcp-write-access-mdbadp1q4 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Claude Code gets auto mode, Figma ships native MCP write access | daily.dev og:description: Anthropic shipped auto mode for Claude Code, a research preview for Team plan users that uses a classifier to automatically allow safe actions and block... og:url: https://daily.dev/posts/claude-code-gets-auto-mode-figma-ships-native-mcp-write-access-mdbadp1q4 og:image: https://api.daily.dev/og/posts/mDBadp1Q4.png og:image:alt: Claude Code gets auto mode, Figma ships native MCP write access og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Claude Code gets auto mode, Figma ships native MCP write access **[Agentic Digest](https://daily.dev/sources/agents_digest)** · 5 min read · 3 upvotes · 0 comments ## Summary Anthropic shipped auto mode for Claude Code, a research preview for Team plan users that uses a classifier to automatically allow safe actions and block destructive ones without requiring constant approval. Figma launched a native MCP write tool letting AI coding agents build directly in Figma files using real design system components, variables, and tokens. Security researchers at RSAC disclosed six CVEs across major AI coding tools including config file injection and a command-swap attack in Cursor, highlighting how these tools have become a significant attack surface due to broad filesystem access. Additional notable items include Claude Code startup time improving 2.8x, JetBrains launching an agentic development platform, Lovable adding built-in penetration testing, Vercel acquiring new.website for v0, and Agoda data showing AI adoption doubled PR merges but nearly doubled review time too. ## Content **TLDR:** Anthropic shipped auto mode for Claude Code, letting the tool decide which actions are safe to run without asking for approval every time. Figma launched a `use_figma` MCP tool that gives AI coding agents write access to real Figma files using your actual design system. Claude Code startup time is down 2.8x and the Agent SDK is 5.1x faster over the last 60 days. Security researchers at RSAC disclosed six CVEs across major AI coding tools, including config file injection and a command-swap attack in Cursor. --- ## Claude Code auto mode Anthropic added auto mode to Claude Code, currently in research preview for Team plan users. It sits between the default (approve every file write and bash command) and `--dangerously-skip-permissions` (approve nothing). A classifier reviews each tool call before it runs, automatically allowing safe actions and blocking destructive ones like mass file deletion or data exfiltration. If the classifier keeps blocking, it escalates to a human prompt rather than failing silently. Enable it with `claude --enable-auto-mode` or through the VS Code extension. Works with Sonnet 4.6 and Opus 4.6. Enterprise and API rollout is coming, desktop app support is still in progress. ## Figma MCP write access Figma shipped a `use_figma` MCP tool that lets AI coding agents write directly to Figma files — not screenshots, not mockups, actual native assets built from your design system. It works with Claude Code, Codex, Cursor, Augment, Copilot CLI, and several others. Agents read your component library first, then build using what already exists: variables, tokens, auto layout. There's also a skills system where a markdown file teaches an agent your team's specific design conventions — no engineering background required to write one. Free during beta, usage-based pricing later. Anthropic and Figma are running a livestream March 31st (9am PST) covering the workflow. ## AI coding tools as an attack surface Check Point's Oded Vanunu presented at RSAC 2026 with a finding worth taking seriously: AI coding assistants like Claude Code, Codex CLI, Cursor, and Gemini CLI have systematically dismantled endpoint security. These tools require broad filesystem access and high privileges, making them invisible to EDR products. His team disclosed six CVEs including MCP consent bypasses, code injection via `.env` and `.toml` config files, and a swap attack in Cursor that replaces approved commands with malicious payloads. The key shift: attackers no longer need traditional malware. A malicious instruction buried in a `.json` or `.env` file is enough. Recommended mitigations include auditing shadow AI usage, sandboxing AI shell tasks, and treating configuration files as code under zero-trust policy. ## Anthropic Economic Index: experienced users iterate, not delegate New research from the Anthropic Economic Index shows how Claude usage changes over time. Longer-term users are more likely to iterate carefully and less likely to hand the model full autonomy. They attempt higher-value tasks and get more successful responses. This tracks with what practitioners report anecdotally — the instinct to just let the agent run is a beginner move. The people getting the most out of these tools are the ones staying in the loop. --- ## Also notable - **Claude Code performance:** Startup time improved 2.8x and the Agent SDK improved 5.1x over the last 60 days, per a post from Jarred Sumner. - **JetBrains Central:** JetBrains announced an agentic development platform for managing fleets of AI coding agents, with early access starting Q2 2026. Integrates with Codex, Gemini CLI, and JetBrains' own tools. - **Lovable adds pentest:** Lovable became the first vibe coding platform with built-in penetration testing, powered by Aikido Security. Checks for OWASP Top 10, privilege escalation, and data exposure. Previously this cost $5k–$50k and took weeks. - **Vercel acquires new.website:** The new.website team is joining the v0 team to build out agent capabilities, with forms, databases, and content management becoming native primitives. - **Qwen 3.5 27B:** Confirmed as the best model to run on a 3090, per community consensus. Kimi K2.5 (1 trillion parameters) was also run locally on an M2 Pro Max as an experiment. - **Google TurboQuant:** Google introduced a compression algorithm that reduces LLM KV cache memory by at least 6x with up to 8x speedup and no accuracy loss. - **OpenAI Sora shutdown:** OpenAI is shutting down the Sora app, with GPU resources apparently redirected toward GPT-5. Codex app discontinuation is also being discussed internally. - **Cursor + Figma:** Cursor can now create new components and frontends directly in Figma using your team's design system, separate from the Figma MCP announcement. - **VS Code weekly releases:** Microsoft shifted VS Code to weekly stable releases. Recent versions add integrated browser debugging, sandboxed MCP server execution on macOS and Linux, and Copilot CLI permission levels for agent autonomy. - **Agoda on AI bottlenecks:** Agoda's engineering team published data showing teams with high AI adoption merged 98% more PRs but saw PR review time jump 91%, arguing that coding was never the bottleneck — specification and verification are. - **DeepSeek still competitive:** One analyst notes DeepSeek V3.2 used roughly 50% of the compute of Kimi K2.5 and GLM-5 and still has the best code pretraining data, calling it surprisingly bullish for the team. - **Cloudflare Dynamic Workers:** Cloudflare launched Dynamic Workers into open beta, letting developers execute AI-generated code in V8 isolate sandboxes that start in milliseconds — roughly 100x faster than containers. Priced at $0.002 per unique Worker loaded per day, waived during beta. --- Tags: [#security](https://daily.dev/tags/security), [#figma](https://daily.dev/tags/figma), [#mcp](https://daily.dev/tags/mcp), [#ai-coding](https://daily.dev/tags/ai-coding), [#claude-code](https://daily.dev/tags/claude-code) [View this post on daily.dev](https://daily.dev/posts/claude-code-gets-auto-mode-figma-ships-native-mcp-write-access-mdbadp1q4) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/claude-code-gets-auto-mode-figma-ships-native-mcp-write-access-mdbadp1q4","headline":"Claude Code gets auto mode, Figma ships native MCP write access","text":"Anthropic shipped auto mode for Claude Code, a research preview for Team plan users that uses a classifier to automatically allow safe actions and block destructive ones without requiring constant approval. Figma launched a native MCP write tool letting AI coding agents build directly in Figma files using real design system components, variables, and tokens. Security researchers at RSAC disclosed six CVEs across major AI coding tools including config file injection and a command-swap attack in Cursor, highlighting how these tools have become a significant attack surface due to broad filesystem access. Additional notable items include Claude Code startup time improving 2.8x, JetBrains launching an agentic development platform, Lovable adding built-in penetration testing, Vercel acquiring new.website for v0, and Agoda data showing AI adoption doubled PR merges but nearly doubled review time too.","url":"https://daily.dev/posts/claude-code-gets-auto-mode-figma-ships-native-mcp-write-access-mdbadp1q4","datePublished":"2026-03-25T04:18:30.002Z","dateModified":"2026-03-25T04:18:49.507Z","author":{"@type":"Organization","name":"Agentic Digest","logo":"https://media.daily.dev/image/upload/s--V91DY4ls--/f_auto,q_auto/v1772617267/logos/agents_digest","url":"https://daily.dev/sources/agents_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/agents_digest","name":"Agentic Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Agentic Digest","item":"https://daily.dev/sources/agents_digest"},{"@type":"ListItem","position":3,"name":"Claude Code gets auto mode, Figma ships native MCP write access"}]} ```