Think inside your AI world.
How do I connect Supabase to Claude?
Supabase's MCP server is organised in feature groups — database, project management, configuration — and its own survey found 29% of startup respondents already running MCP in production.
Direct answer: connect Supabase's MCP server in Claude (Settings → Connectors → authorise; Claude Code via 'claude mcp add'). The surface groups by feature: database access, project management and config. Claude can then inspect schema, query data and manage project settings to the extent you grant — with roughly ten million developers on the platform, this is one of the most-connected servers in the startup stack.
The backend, questionable
What the users table actually looks like, how many rows match a condition, which project settings differ between staging and production — answered against the real backend instead of a mental model of it. For debugging data issues, the live schema read is the whole game.
The safety rule no schema holds
Worked case: you ratified a migration-safety rule — no destructive DDL on production without a restore point verified first, written the day a dropped column turned out to be load-bearing. Supabase executes migrations obediently; 'verify the restore point first' is an operational law about how you use the database, stored in no schema.
Operations, read under the law
Through Unl a migration conversation arrives framed: the proposed change is destructive, and your rule — restore point verified first, remember the column — stands over it before anything runs. The read is one-way; nothing executes through Unl. What changes is that your own operational law is in the room when the dangerous suggestion sounds convenient.
Supabase to Claude reads and manages the backend; Supabase through Unl puts your operational laws in front of every backend conversation — the restore-point rule arrives before the destructive command does.
Reads through Unl arrive with measured context — in the presence of the decisions you’ve already settled. The reach lane is live: one box, paste anything. If it speaks MCP, Unl can reach it. Readings arrive unprompted, the data beside the criterion; Unl is a courier, not a warehouse, and keeps only your keys and the frame.
Read further
Questions people ask
What feature groups does the Supabase MCP server expose?
Database (schema, queries), project management and configuration — grouped so you can grant the slice you are comfortable with. Reads are the conservative default; writes and config changes deserve deliberate scoping.
Is Supabase-over-MCP common in production?
Supabase's own State of Startups found 29% of respondents already using MCP in production and 28% experimenting — unusually high adoption, which matches how naturally a backend pairs with a conversational client.
Does Unl execute anything against my database?
Not while measuring — a read only reads, so checking a migration against your ratified rules never runs anything. The honest answer past that is yes, it can: where a server exposes a tool that executes, Unl can call it, because the socket exposes a server's full surface rather than a chosen subset. That call carries your explicit in-turn gesture and is verified at the choke point, and a tool that does not declare itself read-only is treated as a write by default rather than assumed safe. Nothing runs as a side effect of a read; it runs when you ask and confirm.
Does Unl keep a copy of my data?
No. What Unl reads is quarantined for the turn and never kept — no query text and no read content reaches our telemetry. Supabase stays the record. What Unl holds is the rule you ratified and its reasoning, which is a few lines you wrote, not a mirror of your database.
What this is
Think inside your AI world — you stay in command
Save the thoughts, decisions and targets worth keeping, each with its reasoning, carried into every AI session the moment they matter. A new unit of exchange between you and your AI: the Settled Why with standing that travels. Unprompted.
Your whole AI world. What you decided at the epicentre. It plugs into Claude, Claude Code, ChatGPT and Cursor as an MCP connector — quick to connect, in a couple of steps.
MCP native·Human settled·Model agnostic·Your data
Measured Context
Connect Unl to bring the right information into the moment.
Your sources, read against the criteria you set.
The full product, open. Free at launch.