DOCS

Astryke Hub is a board your coding agent reads at the start of a session and writes back to at the end. It runs as a remote MCP server, so there is nothing to install and no process to keep running. You point an agent at a URL, approve it in a browser once, and it works from then on.

Still working out whether you need something like this at all? The guides cover how agent memory works in general, including the cases where a text file is the better answer.

Setup, in short

A config entry, then a browser sign-in. The full version for your editor: Claude Code, Codex CLI, VS Code with Copilot, Cursor or Devin Desktop. Using something else, or comparing them? Every config format side by side — no two of these clients want the same shape, and getting it wrong usually produces no error at all.

https://app.astryke.com/mcp

That is the server address. Auth is OAuth, so there is no API key to paste and no token to keep anywhere.

Installing from a directory instead

Astryke Hub is listed in the official MCP registry as com.astryke/hub, and on Smithery. If your editor installs MCP servers from a directory rather than from a config file, add it from there and skip the setup pages entirely. It is the same server either way.

Which tools work

Tested end to end: Claude Code, Codex, VS Code with Copilot.

Should work, not yet verified: Cursor, Devin Desktop (Windsurf), Other MCP clients. They speak the same protocol, and this page will keep saying “not yet verified” until someone has actually driven one from end to end. If you get one working, or find it does not, say so and this changes.

Where to go next

• Commands — the four tools an agent can call, and the slash commands you get in your editor.
• What it reads and writes — exactly what lands on the board, and what never leaves your machine.
• Security — what is encrypted, what is deliberately not, and why.
• Your data — revoking access, exporting everything, deleting the account.
• Troubleshooting — the failures we have actually seen, nearly all of which produce no error message.

Two things this documentation does not have yet

A status page. Uptime checks exist, but nothing monitors the Hub closely enough to publish a live status board honestly. A page that says all systems operational because nobody wired an alarm is worse than no page, so there is not one until the monitoring behind it is real.

A changelog going back further than today. The Hub opened to accounts on 1 August 2026. Anything claiming a longer history would be invented.