One command captures your entire project state. Switch between Claude Code, Cursor, Copilot, Codex — the next AI picks up instantly. Now with native MCP integration.
mindswap maintains a universal handoff document. Any AI reads it instantly and knows exactly where you left off.
mindswap — everything auto-capturedmindswap captures eight types of context — mostly automatically.
Branch, diff summary, staged files, recent commits. Auto-captured on every save.
30+ frameworks, 15+ databases, monorepo tools, CI/CD, Docker, Kubernetes. Auto-detected on init.
Auto-detected from branch name and commits. Or set manually with blockers and next steps.
WHY you chose something. Auto-warns if a new decision contradicts an old one.
New packages auto-logged as decisions. Removed deps tracked too. Compares against your decision log.
Run tests during checkpoint. Next AI knows "47 tests passing" or exactly what's broken.
.mindswap/A single directory at your project root. HANDOFF.md and state.json are committed so your team benefits. History and branch states stay local.
The entire flow. Everything else is automatic.
Auto-detects your stack, imports existing AI context files (CLAUDE.md, .cursorrules), installs git hooks for auto-tracking.
npx mindswap initOne command. Auto-detects task from branch name, captures git state, logs dependency changes, generates all context files.
npx mindswapArchive the task, reset to idle. Your decisions and history stay. Start the next feature with a clean slate.
npx mindswap doneSmart features that make every AI handoff seamless.
State auto-saves per git branch. Switch to feat/payments and it loads that branch's task, decisions, and checkpoint. Switch back, your main state is restored.
Log "NOT using Redis" then later "using Redis"? mindswap warns you. Also catches reversed choices ("chose X over Y" then "using Y") and package.json contradictions.
Added Stripe to package.json? mindswap auto-logs "added Stripe (stripe@^14)". Removed Redis? Logged too. No manual decisions needed for dep changes.
Already have a hand-written CLAUDE.md? mindswap appends its section inside markers. Your content is always preserved. Works for AGENTS.md and copilot-instructions too.
mindswap switch cursor — saves state, generates Cursor-specific context, opens Cursor. Works for Claude, Copilot, Codex, Windsurf.
Full session narrative: task, commits, decisions, conflicts, tool usage chart. Machine-readable JSON output for CI/CD integration.
The core flow is 3 commands. The rest are power-user tools.
mindswap and switch.Native context files for every major AI coding tool. Safe merge — your content is never overwritten.
Not just static files. AI tools call mindswap directly via Model Context Protocol. One call = full project briefing.
"What do I need to know?" — Called at session start. Returns synthesized project state: task, decisions, conflicts, test status, recent work. One call replaces reading 5 files.
focus: all | task | decisions | recent
"Here's what I did." — Called at session end. AI saves its summary, decisions, next steps, blockers. The next session (in any tool) picks up seamlessly.
summary + decisions + next_steps
"What did we decide about auth?" — Called mid-session. Natural language search across all decisions, history, and project state.
query: "database choice"
Two commands to set up everything — no global install needed:
npm install mindswap --save-dev
npx mindswap mcp-install
Only 3 tools. Research shows AI accuracy drops from 82% to 73% past 20 tools.
We chose quality over quantity.
Every context file is scanned for API keys, passwords, tokens, and connection strings before writing. Secrets are auto-redacted.
AWS keys, GitHub tokens, Stripe keys, OpenAI keys, database URLs, JWT secrets, private keys, bearer tokens — all detected and redacted before any file is written.
mindswap pr adds context summary to GitHub PRs. Task, decisions, test status — auto-injected with safe markers so reviewers get instant context.
Detects Turborepo, Nx, Lerna, npm/yarn/pnpm workspaces. Tracks state per package. Shows which packages have changes in HANDOFF.md.
Grades your handoff context A–F (0–100). Tells you exactly what's missing: "no decisions logged", "checkpoint stale", "no test status". Improve as you go.
Install mindswap in 10 seconds. Native MCP for Claude Code, Cursor, and VS Code.