Stop re-explaining your project every AI session

The problem with AI IDEs today

AI coding tools like Cursor, Claude Code, and VS Code Copilot are powerful — but each new chat starts from scratch unless you've manually set up a rules file. Even then, your rules file doesn't capture decisions made last week, or what you were building yesterday. Your AI has no automatic access to:

• the decisions you made last week ("we chose Drizzle over Prisma because of better edge support")
• the conventions you care about ("no semicolons, functional components only, strict TypeScript")
• what you were building yesterday ("we were halfway through the auth flow")
• your stack ("Next.js App Router, Neon, Better Auth, pnpm workspaces")

Rules files (.cursorrules, AGENTS.md, CLAUDE.md) are a good start — and Memcone reads them. But static files don't capture decisions made since you last edited them, and they don't include a summary of what you were building. You end up re-typing context at the start of each session anyway.

How Memcone fixes it

Memcone stores your project's detected stack, rules files verbatim, architectural decisions you log, and session summaries your AI writes — then injects a compiled version at the start of every AI session via MCP (Model Context Protocol).

This is retrieval, not cognition. Memcone doesn't make your AI smarter — it makes your project state available at session start so you don't have to paste it in manually.

## Project Identity
language: TypeScript
framework: Next.js App Router
packageManager: pnpm
semicolons: false
strictMode: true

## Hard Rules
[verbatim content of your AGENTS.md / .cursorrules]

## Architecture
auth: Better Auth
orm: Drizzle ORM
database: Neon Postgres
deployment: Vercel

## Key Decisions
- Chose Drizzle over Prisma for better edge runtime support
- Redis for session caching, not database sessions
- Stripe metered billing, not seat-based

## Recent progress
[what you built last session, what's in progress, what's next]

Two layers work together. Your project state (stack, rules, decisions) is set once with the CLI and updated on demand with npx @memcone/cli sync.Session notes (checkpoints) are stored when your AI calls memcone.remember — retrieved at the start of the next session via memcone.context. Nothing is captured without a tool call.

Set up in 3 steps

npx @memcone/cli init
npx @memcone/cli link

// Cursor — .cursor/mcp.json
// Claude Code — claude_desktop_config.json
// VS Code — .vscode/mcp.json
{
  "mcpServers": {
    "memcone": {
      "url": "https://memcone.com/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_MEMCONE_API_KEY"
      }
    }
  }
}

Capturing what you built — the checkpoint

Project identity and rules are always present. Capturing what you built last session usually means asking for a checkpoint so the model calls memcone.remember. Decisions and preferences you state during the session are often stored the same way as the chat goes — no checkpoint required for those.

For session-level continuity, end your sessions explicitly:

you:  "checkpoint"
  or: "save our progress"
  or: "wrap up and store what we did"

AI:   calls memcone.remember →
      "Built auth flow with Better Auth. Decided against Clerk
       due to cost at scale. Email verification in progress.
       Next: add Google OAuth and magic link providers."

The next session loads that summary through memcone.context. Your AI picks up where you left off — without you re-pasting context.

What gets auto-detected

npx @memcone/cli link scans your repo and automatically picks up:

Local skills (from .agents/skills/, .claude/skills/) are also synced automatically and injected via the same MCP channel.

Supported AI tools

Any MCP-compatible AI tool works. Tested integrations:

Store your project state once. Load it everywhere.

10,000 free units. No credit card. Works with Cursor, Claude Code, VS Code, and any MCP-compatible tool.