Faster path from Cursor to Claude Code
One Click merges MCP from both Cursor and VS Code/Copilot into Claude Code's native config (~/.claude.json) with case-insensitive dedup. Sync rules and scaffold a memory bank — all in-process, no npx needed.
One place for Claude Code-related setup.
Stop losing time to scattered rules, MCP config, and “skills” folders. Give your team a single, honest view inside VS Code and JetBrains (IntelliJ, PyCharm, …). Both IDEs ship Agentic Teams — 8 collaboration protocols, agent/team/command CRUD, live transcript, and approval gates — from the same repository.
After you install: In VS Code, use the Activity Bar (icons on the
far left) → click Claude Code ToolBox → in the Side Bar click
MCP & skills. That opens the hub (Intelligence, MCP, Skills, Workspace). Reload the window
if you don’t see the icon yet.
JetBrains (IntelliJ IDEA, PyCharm, …): install Claude Code ToolBox from the
JetBrains Marketplace,
open in your IDE, or read the
IntelliJ plugin readme (MCP hub + Agentic Teams with all 8 protocols).
In plain terms: Claude Code in VS Code only works as well as the setup around it—but that setup is
usually scattered across files, machines, and habits.
Claude Code ToolBox is one dedicated toolbox in VS Code: see
what’s configured, standardize how teams adopt Claude Code (including from Cursor and optional GitHub Copilot migration steps), and build explicit context while each developer still chooses what to share.
For engineering teams
Discover and add from one hub
Browse MCP registry and skills catalogs, install servers directly to ~/.claude.json or .mcp.json with a scope picker — fewer raw JSON edits by hand.
One checklist view
Workspace vs personal setup, local skill folders, rules, CLAUDE.md, memory bank—aligned with what you shipped.
Smarter context for Claude Code
Context packs and readiness flows with explicit choices so teams agree on what Claude Code is allowed to see.
AntiVibe Safety Guards — destructive commands + domains + supply chain
Destructive Command Guard blocks rm -rf, git push --force, git reset --hard, DROP TABLE, curl | sh, and 30+ patterns before they execute. Domain Whitelisting enforces an allowlist for all web requests — prevents data exfiltration to unknown hosts. Supply Chain Guard blocks installation of known-compromised packages (event-stream, node-ipc, colors, faker, and more). Configurable patterns, overrides, block/warn modes. Installed by default via One Click Setup.
Agentic Teams — skills, memory, swarm dispatch + multi-agent debate
10-agent SDLC starter pack (incl. UI/UX Designer) that debate a design and produce a plan you approve before code runs. One-click install enables CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS and writes agents/teams to ~/.claude/ for native discovery. Reset button removes everything in one click. Agent Dashboard shows live cost estimates for all sessions. Skill-backed agents point to SKILL.md instead of freeform prompts. Per-agent long-term memory lets agents learn and retain knowledge across runs. Eight collaboration protocols, swarm dispatch, 7 preset teams. Full parity on VS Code, IntelliJ, macOS, Windows, and Linux.
One Click Setup & Thinking Machine Mode
Two highlighted controls on the Intelligence tab turn messy, one-off Claude Code setup into something your team can repeat and reason about.
One Click Setup
What it does: After you confirm you understand the risk, it runs the sequence you configured in Settings — typically migration tracks for Cursor → Claude Code and optional GitHub Copilot → Claude Code: port Cursor MCP into VS Code
mcp.json, Claude-oriented memory bank init, Cursor rules into CLAUDE.md, optional .cursorrules merge, skills
migration, MCP & Skills awareness (under .claude/) with optional merge into CLAUDE.md, readiness summary, config scan, and follow-ups. Bridges use the
bundled Node CLIs shipped with the extension (no npx fetch).
Why it matters: Onboarding and “make this repo Claude-ready” shouldn’t depend on who remembers which README or which CLI flag. One Click encodes your team’s playbook once, then anyone can execute the same steps and review the same terminals and files.
Note: After One Click Setup completes, you’ll be prompted to close and reopen VS Code. Hooks (Token Optimization) require a fresh window to take effect. AntiVibe Safety Guards must be enabled separately via the hub toggle.
Thinking Machine Mode
What it does: A master switch for session priming — optional MCP & Skills awareness (writes under .claude/) plus optional merge into CLAUDE.md, and a
context pack for Claude Code, with settings for confirmations and defaults. The first time you enable it, VS Code asks you to
Engage (or cancel to turn the mode off again). Turning the mode off clears that acknowledgment so the next enable shows the dialog again.
Why it matters: Claude Code works best when sessions start from accurate, explicit context — what MCP servers exist, where skills live, and a structured snapshot of the workspace — not from guesswork. Thinking Machine Mode makes that priming a deliberate, repeatable gesture instead of a forgotten manual step.
Agentic Teams — multi-agent planning & debate
The 🤝 Agentic Teams tab (now the second tab for faster access) turns the hub into a multi-agent planning & debate workbench. Agents can be backed by SKILL.md files and have per-agent long-term memory that persists learnings across runs (with a global toggle). Specialised agents argue a design, produce a plan you approve, then execute — with a live transcript and persisted artifacts. All sections are collapsible. Everything is opt-in, disclosed upfront, and reversible.
Agent Teams + skill-backed agents + long-term memory + 8 protocols
What it does: CRUD for Claude Code subagents (YAML-frontmatter .md under ~/.claude/agents/). Skill-backed agents: point any agent to a SKILL.md instead of a freeform prompt (radio: Custom prompt vs Use skill with dropdown of all discovered skills). Per-agent long-term memory: agents learn from interactions and retain knowledge across runs as <agent>.memory.md alongside the agent file; a global toggle bulk-enables memory for all agents. A 10-agent SDLC starter pack (incl. UI/UX Designer for Figma review and component layout) and 7 preset teams land with one click. Default model: (inherit caller default) so agents use whatever model your session is configured with (Opus 4.7, Sonnet 4.6, Sonnet 4.5, or Haiku 4.5). Eight collaboration protocols drive runs: native-task, round-robin, handoff, orchestrator, parallel-fan-out, debate + judge, plan-then-code (with your approval gate), and converge (parallel → cross-pollinate → synthesize → approve → execute). A live color-coded transcript shows pulsing status, tokens, cost, projection, and an approve-plan modal.
Why it matters: Agents that remember what you taught them last time and can use curated skill instructions are far more effective. Planning and reviewing benefit from disagreement — debates and plan-then-code flows make multiple perspectives explicit and auditable, with JSONL transcripts per run.
Agent Dashboard — live visibility for every Claude Code session
What it does: A strip of kanban cards above the Run panels shows every running Claude Code session — ours or one you started in a terminal or another window. Each card surfaces pulsing status, current tool + target file, context-window fill, tokens in/out, and estimated USD cost (token-based cost estimation using Sonnet pricing when the CLI doesn't report cost directly). Last-3 tool-call feed and an inline "needs approval" badge. Phase 2 adds swim-lane grouping by workspace, search/filter, cost-cap warnings (soft breach → Stop-now toast; hard breach → auto-abort), and foreign-hook detection.
Why it matters: Strict opt-in. On enable, the Toolbox drops a Python hook helper into ~/.claude/, registers 5 events in ~/.claude/settings.json (atomic, dedup, reversible), and starts an HTTP listener on 127.0.0.1:3456. No telemetry — nothing leaves your machine. Disable reverses only the Toolbox's entries and deletes the helper.
Swarm dispatch — teams are slash commands
What it does: Every team is a slash command. Creating or editing a team auto-generates a /command that dispatches all agents in parallel via the Task tool (swarm pattern), then synthesizes results. 7 preset teams ship with the starter pack:
/debate-team <topic>— product-manager, architect, security-reviewer in parallel → synthesis./plan-team <task>— product-manager + architect parallel planning → unified plan./review-team— code-reviewer + security-reviewer parallel review → merged findings./security-team <change>— OWASP threat model./sdlc-plan-then-code <task>— full SDLC swarm./refactor-team <target>— backend + frontend + QA + reviewer parallel refactor./spec-team <idea>— PRD + technical addendum in parallel.
Why it matters: One concept, one place. Create a team in the UI and it's immediately available as a swarm /command in any claude session — from VS Code, JetBrains, a terminal, or any other IDE. No separate command editing needed. Cross-platform: Windows, macOS, Linux.
AntiVibe Safety Guards — protect against destructive commands, data exfiltration & supply chain attacks
Claude Code runs arbitrary shell commands, fetches URLs, and installs packages. AntiVibe Safety Guards installs lightweight Python hooks into ~/.claude/settings.json that intercept dangerous operations before they execute.
Destructive Command Guard
What it blocks: rm -rf, git push --force, git reset --hard, git branch -D, git clean -fdx, DROP TABLE, TRUNCATE, curl | sh, chmod 777, dd if=, npm publish, and 30+ more patterns. Mode: Block (exit 2, prevents execution) or Warn (stderr only). Allow overrides: permit specific patterns when needed (e.g., rm -rf node_modules). All configurable in VS Code settings.
Domain Whitelisting
What it does: Enforces an allowlist (or blocklist) for all web requests. Default allowlist includes GitHub, npm, PyPI, MDN, Stack Overflow, Read the Docs, and 30+ trusted documentation sites. Unknown domains are blocked (exit 2). Prevents prompt-injection-driven exfiltration to pastebins, file-sharing sites, or attacker-controlled hosts. Add your internal domains with wildcard support (*.internal.company.com).
Supply Chain Guard
What it blocks: Installation of known-compromised packages: event-stream, ua-parser-js, colors, faker, node-ipc, peacenotwar, es5-ext. Intercepts npm install, pip install, yarn add, pnpm add, gem install, cargo add, and bun add. Mode: Block (exit 2, prevents install) or Warn (stderr only). Customize the blocklist in VS Code settings.
Triple-confirmation flow: In block mode, all guards use a 3-step confirmation. 1st attempt blocks silently. 2nd attempt tells Claude to ask you for permission. 3rd attempt proceeds only after you confirm. Counter resets after each allow cycle — next occurrence starts fresh. Session-scoped, expires after 1 hour.
Setup: AntiVibe Safety Guards must be enabled separately via the hub toggle or command palette (not auto-enabled by One Click Setup). After enabling, close and reopen VS Code for hooks to activate. Fully reversible via Disable.
What you get in VS Code
First: after install, open Claude Code ToolBox in the Activity Bar (left edge), then MCP & skills in the Side Bar—that is the hub. Below: access walkthrough, then high-resolution hub captures. Click any image for fullscreen.
Ready in the Marketplaces
VS Code: search for Claude Code ToolBox (MCP, Skills, Cursor/Copilot → Claude) or use the button below. After install, open Claude Code ToolBox in the activity bar, then MCP & skills.
Install in VS CodeVS Code did not open? Open the VS Code Marketplace in your browser instead.
code --install-extension amitchorasiya.cloude-code-toolbox-vscode
JetBrains (IntelliJ IDEA, PyCharm, …): install Claude Code ToolBox from the
JetBrains Marketplace or use
jetbrains://… to open your IDE’s plugin dialog (plugin id
com.amitchorasiya.claude.code.toolbox). Preview builds: sources & Gradle.