Less muri. More mural.

Data is a wall
everyone can paint.

Muri lets your staff build their own apps and dashboards with Claude, in the tools they already use, and keeps the data behind them in one governed place.

Muri is the Lean word for overload; we help you turn it into structure. Every layer is kept, every change says who made it, and the wall's owners pick who paints.

Grab a can How it works
keep this week's support tickets so I can chart them

Claude used create_dataset and insert

support-tickets · ds_k3v7qd2m5xw4ha6n owner you version 1 · 128 documents
what changed since Monday?

Claude used history

Three changes: Maria closed 9 tickets, 14 new ones came in, and one was reopened. Want the chart updated?

Just say it

Your agent does the database work. Data lives in datasets: collections of JSON documents in whatever shape suits your app. Behind each sentence, Claude calls a tool on your behalf, and Muri checks your role on every call.

"store the sales targets from this sheet" "how many orders per week and region?" "what did Simon change yesterday?" "let everyone here read the dashboard data" "make Anna an owner too" "archive last year's campaign"

Dashboards

Query, count, sum and average your documents, grouped however the chart needs them.

Apps

Artifacts and pages built with Claude read the same data, as structured answers they don't have to parse.

Syncs

Pull from another system on every refresh: many writes in one call, all or none, never a duplicate.

Crews

Every team gets a wall. Share a dataset by email, as viewer, editor or owner, or with everyone here.

Paint with data. Keep every layer.

What changed since Monday?

History lists every change after a version you name: who, when, and the patch to each document. Earlier versions stay readable for a year.

No paint over paint

A write can say which version it read. If someone changed the document since, nothing is written, and the agent reads again.

It explains itself

Owners describe what each dataset and field means, so the next agent understands the data. A refusal names its rule and how to fix it.

insertversion 1updateversion 2

Your people keep working in Claude as they do today. Claude connects to Muri over MCP, they sign in with their Google account, and each write is recorded with who made it and exactly what it changed.

Data's for everyone. The owners pick who paints.

Sign in with Google

Click, choose your account, approve. Claude gets its keys by itself; nobody handles a password.

Invite-only

An account that isn't on the list is turned away before anything is issued.

Roles per dataset

Viewers read, editors write, owners describe and share. A dataset you have no role on looks like one that doesn't exist.

A record that stays

Every grant and revoke is logged for good. Every write is kept, with who made it, for a year.

Undo for a year

Archive makes a dataset read-only; a deleted one can be restored, with its history, for a year.

No new tool to learn

Everyone keeps working in Claude: on the web, the desktop, the phone or in Claude Code. Muri is one connector.

Next on the wall

Not built yet. Where Muri is heading, so your data stays yours wherever it lives:

Grab a can. Bring your agent.

This Muri is invite-only: ask the people who run it to add your Google address. Then connect Claude with this address:

https://mcp.getmuri.dev/mcp

claude.ai, desktop and phone

  1. Settings → Connectors → Add custom connector.
  2. Paste the address above and click Connect.
  3. Sign in with Google and click Allow.

Claude Code

  1. Run claude mcp add --scope user --transport http muri https://mcp.getmuri.dev/mcp
  2. In Claude Code, type /mcp, choose muri, Authenticate.
  3. Ask: "create a dataset for my reading list".

This is a playground with no backup yet, so keep test data here.