One library of skills & memories, shared by every agent.

DexDev stores your agent skills and memories once and serves them over remote MCP to Claude, ChatGPT, Codex, Cursor, Gemini CLI and any other client you connect, with permissions you grant and revoke per client.

Free plan, no card needed, and no AI keys to manage: your own AI clients do the work.

code-reviewprefers pnpmprefers pnpmrelease-notescode-reviewdecision: use Postgresprefers pnpmdeploy-checklistcode-reviewtone: concisecode-reviewprefers pnpmcode-reviewprefers pnpmcode-reviewDexDevskills · memoriesClaudeChatGPTCodexCursorGemini CLI
SkillMemorySame library, every client
Works withClaudeChatGPTCodexCursorGemini CLIany remote MCP client

How it works

Teach one agent. Every agent knows.

Three steps, then your context follows you from client to client.

  1. 1

    Store once

    Keep SKILL.md folders with their scripts and references, plus memories: preferences, facts, decisions and notes. Or ask your local agent to upload everything it already has and triage it into one clean library.

    code-review/
      SKILL.md
      scripts/lint.sh
      references/style.md
  2. 2

    Connect clients over MCP

    Add one MCP URL to each AI client, sign in with OAuth and approve exactly what that client may read or write. Revoke it whenever you like.

    Claude   read skills · read memories
    Codex    read skills · write memories
    Cursor   read skills
  3. 3

    Every agent, same context

    Your agents search the same published skills and active memories, and fetch bundled files only when they need them. Teach one agent, and the rest already know.

    > search_skills "review"
      code-review · v4 · global
    > search_memories "package manager"
      prefers pnpm · preference

Features

Everything your agents should share, in one place

Built for people who use more than one AI client and are tired of repeating themselves.

  • Skill folders with scripts

    Each skill is a real folder: SKILL.md plus scripts, references and assets. Agents load the instructions first and fetch bundled files on demand. Import and export as Markdown or ZIP.

  • Memory folders

    Preferences, facts, decisions and notes, organized in folders. Memories written by an agent wait as pending until you confirm them.

  • Multi-project scopes

    Global records apply everywhere; project records stay inside their project. Search respects scope and works across languages.

  • Per-client permissions

    Grant read skills, read memories or write memories per client. Enforced in the database and revocable in one click.

  • Version history & publish

    Drafts, publishing, full history and restore. Agents only see the skills you publish.

  • In-app assistant

    Draft a new skill or tidy an old one with the built-in assistant. Nothing is published without your review.

  • Import from your local agent

    Ask Claude Code, Codex or any local agent to upload every skill it has. Duplicates and near-duplicates are triaged into one clean library instead of being copied by hand.

Pricing

Start free. Upgrade when your library grows.

Prices in US dollars. Every plan works with every MCP client.

Free

$0forever

Everything you need to give your agents one shared library.

  • 50 skills
  • 200 memories
  • 3 projects
  • 20 assistant messages per day
  • Every MCP client, per-client permissions
  • Full export, any time
Start free
No limits

Pro

$8per month

For people whose agents do real work every day.

  • Unlimited skills
  • Unlimited memories
  • Unlimited projects
  • 300 assistant messages per day
  • Everything in Free
Go Pro

Trust

You stay in control

DexDev is a library, not a runtime. Your agents only get what you hand them.

  • Scripts never run here

    Bundled scripts are stored and served as files. DexDev never executes them.

  • Your data stays portable

    Export skills as Markdown or ZIP and your whole account as JSON, any time.

  • You decide per client

    Choose what each client can read or write, and revoke access in one click.

Does DexDev run the scripts bundled with my skills?

Never. Scripts, references and assets are stored and served as files. Your agent decides whether to run anything, inside its own environment and under its own permissions.

Who can read or change my library?

Only the clients you approve. Each client gets its own permissions (read skills, read memories, write memories) on a consent page, enforced in the database. Revoke a client and it loses access immediately. Memories an agent writes stay pending until you confirm them.

Can I take my data with me?

Yes. Export a single skill as Markdown or a ZIP folder, or export your whole account as JSON whenever you want.

Do I need a model API key?

No. Inference happens in the AI clients you already use. DexDev only stores and serves your skills and memories.

Does it replace my client's built-in memory?

No. DexDev is an external library your agents call over MCP when they need it. It works alongside whatever a platform already remembers.

Give your agents one shared memory.

Sign in with your email, connect your first client and import the skills you already have.