Skip to main content
Glama

All-Skill

One memory and skills hub for every AI app you use. Claude and ChatGPT each keep their own memory and their own skills, so switching between them means re-explaining who you are, what you're working on and how you like things done. All-Skill gives them one shared place instead: your profile, your memories and your skills, stored as plain Markdown files and served to each app over MCP.

  • Every app reads and writes the same hub. A fact you tell ChatGPT is there when you open Claude.

  • Each app gets the hub the way it likes it. Claude gets a MEMORY.md-style index and loads skills by name, mirroring its own system; ChatGPT gets OpenAI-style search/fetch. The data underneath is identical.

  • Skills in the open SKILL.md format that Claude and ChatGPT both use, with supporting files.

  • Safe writes across apps: every memory has a version, so one app can't silently overwrite another, and corrections update the existing memory instead of piling up duplicates.

  • A saving policy that travels with the hub: what apps may save, and what needs your OK first, such as health or finances.

  • Full history: every change is a git commit credited to the app that made it, viewable and restorable in the web UI.

  • Web UI to browse, edit and audit everything, including a "see what Claude / ChatGPT sees" preview and an activity log of every tool call.

  • Google sign-in (OAuth 2.1 with PKCE and dynamic client registration) when hosted. No dependencies, Node 20+.

Set it up with your AI agent

Paste this into Claude Code, Codex or any coding agent:

Set up All-Skill for me: https://github.com/shai840/all-skill-hub

Clone it and follow SETUP.md step by step. Ask me whether I want hosted (Railway, works in Claude web/desktop/mobile and ChatGPT) or local (Claude desktop and Claude Code on this computer only). Tell me exactly which accounts and keys I need before you start, do everything you can yourself, and stop and walk me through any step only I can do (Google Cloud Console, pasting instructions into app settings, signing in). Never print or ask me to paste secrets into the chat. At the end, install the "use the hub" instructions where you can, give me the ones I have to paste myself, and run the final check in SETUP.md.

Doing it by hand? SETUP.md works for people too.

Related MCP server: .memex

What you'll need

Hosted

Local

Node 20+, git

✓

✓

Railway account + CLI

✓

Google Cloud OAuth client (Web application). An existing one can be reused.

✓

Private GitHub repo + token for history backup

optional

Connect your apps

Once it's running, connect each app to the hosted endpoint: https://YOUR-HUB/mcp in Claude and Codex, and https://YOUR-HUB/mcp/chatgpt in ChatGPT.

ChatGPT custom connector/plugin: Create the custom MCP plugin in ChatGPT and connect it to Railway. The current ChatGPT screen calls these Plugins: open ChatGPT Plugins → Add → Add custom MCP server. The full connection guide also covers Codex, OAuth sign-in, checks, and troubleshooting.

Make your apps actually use it

Connecting the tools isn't enough. Each app needs an instruction to use the hub for normal recall and deliberate saves:

Then review existing memories with the bundled import-native-memory skill. Keep each app's native memory enabled as a backup, and use the hub for normal recall and deliberate saves. Some native details may not be exportable; the backup can help recover information missed or deleted from the hub. docs/testing.md checks the connection and behavior.

Where your data lives

The code repo never contains your data. hub/ is git-ignored.

  • Hosted: on the server's persistent volume, as Markdown files with their own git history. Set DATA_REPO_URL to push that history to a private GitHub repo as an off-site backup.

  • Local: in ./hub, created from templates/hub on first run.

Develop

npm start        # local hub + web UI at http://localhost:4747, reloads on code changes
npm test         # store, adapters, stdio/HTTP transports, OAuth flow, history
npm run deploy   # tests, then `railway up`

How it fits together, and how to add an adapter for another app: docs/ARCHITECTURE.md.

License

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables Claude and any MCP client to store long-term memory as plain Markdown in the user's own private GitHub repo, with hybrid semantic search, a knowledge graph, auto-memory, and commit history. It also keeps working state alive across context compaction through core memory, hook-driven checkpoints, and a resume tool available in any client.
      1
      MIT
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI coding agents to persistently remember and recall insights across sessions as atomic Markdown knowledge cards with bidirectional links, shared across Claude Code, Cursor, VS Code Copilot, Codex, Windsurf, and any MCP client. Stores memory as plain markdown files with git sync instead of a vector database, letting agents build on prior knowledge rather than starting from scratch.
      9 npm
      ISC
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables people and AI assistants to share one persistent Markdown-in-Git memory, so any MCP client can load current context, save sessions and decisions, search notes, and work from a team's real knowledge across machines.
      2 npm
      AGPL 3.0
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables Codex and Claude to share Git-backed Markdown memories, assistant handoffs, and reusable skills through read-only and write-enabled MCP endpoints, while owners review revisions, conflicts, and deletion requests in a private GUI.
      MIT