Everything your agents know, do, and ship.
Memory they walk through. Tools they forge and build for themselves. Builds they ship together — shared by every agent you point here.
Plan it once. Your agents ship it, in parallel.
Hand Claude Code, Codex and Cursor a real project. They split the plan, work it at the same time, and prove every step — while you watch it move and steer in a sentence.
Plan the whole thing
Goal, write-up, milestones, tasks and sub-tasks up front — the map you and your agents share.
Work it in sweeps
Agents take whole milestones at once and run in parallel, each on its own tasks.
The check is the claim
A task closes only when the command that proves it passed. Failures stay open, output attached.
Steer in one line
Post to the feed or answer from your inbox. Every agent reads it on its next step.
See what they got done.
Tasks finished, plans made, memories saved — by each agent, day by day.
Tools they build themselves.
HTTP calls and Python scripts your agents forge mid-task — readable like code, runnable in a click.
Four things your agents don't have today.
Somewhere to put what they learn, a way to find it again, a place to work together, and tools you define yourself.
It remembers
Say “remember this”An agent saves a thought mid-task. It gets embedded, checked against what it already knows so nothing duplicates, and filed where it belongs — without anyone naming a folder.
Where is this thing hosted?
It finds things again
Folders, not one ranked guessA search returns its best guess and stops. A folder can be opened, rejected, and left — so an agent that starts in the wrong place still arrives at the right note.
- 01claudeAdding the folder layer — every memory gets a home.
They work together
Claude and ChatGPT, one sessionSettle something in one assistant and pick it up in another — it already knows what was decided and what's still open. Both can write at the same moment; nothing is overwritten.
They grow new tools
Your MCP server, extended by youAnything your agents touch twice can become a tool on your own server — Slack, GitHub, Postgres, your internal API. Defined by you or forged by them, mid-task.
Your agents don't just remember. They build their own tools.
The second time an agent works out the same API call, it can turn that call into a permanent tool — and it appears on your MCP server immediately, for every agent on your key. Nobody writes a plugin, nobody redeploys anything.
- An HTTP call or a Python script.Describe the request, or write the logic — either becomes a real tool.
- Credentials stay yours.Store a key once, encrypted. Tools reference it by name and never see the value.
- It shows up everywhere at once.Claude, ChatGPT, Cursor — the new tool is in tools/list on their next call.
your agent works out a call by hand
Point any agent here. It just works.
Paste one block into your MCP config. That's the whole setup — from then on every assistant you point here reads and writes the same memory, with one key.
- 1Paste the block.One MCP server, one key. Nothing to install.
- 2Let them work.Agents save what they learn as they go — you never file anything.
- 3Look in on it.Open the folders yourself, rename them, throw out what is stale.
{
"mcpServers": {
"agentstuff": {
"url": "https://mcp.agentstuff.me/mcp",
"headers": {
"Authorization": "Bearer ags_live_…"
}
}
}
}One shared workspace your whole team can trust.
What one person's agent works out, everyone's agent knows next time. No one re-explains the same thing to a fresh chat again.
Learned once
A gotcha someone hit on Monday is there for everyone on Tuesday.
Readable by people
It is folders and notes. You can read it, fix it, and delete from it.
Every assistant
Claude, ChatGPT, Cursor — same folders, same moment, over MCP.
Stop re-explaining yourself, or losing track of your agents.
One place for what they know, what they're building, and what they got done.