MCP · Claude Code · Cursor · VS Code

Your AI agent wants to change the database. Make it open a PR first.

Connect Claude Code, Cursor, VS Code, or any MCP-capable agent to nuzur. Every schema and data change the agent makes becomes a change request — a reviewable diff you approve before anything touches your database.

Connect your agent — free Watch the 1-minute demo ↓
The problem

"It ran the migration. On prod."

AI agents are shockingly good at writing SQL. They're also confident, fast, and have no idea which table is load-bearing.

Right now your options are bad:

Give the agent write access

…and hope the prompt was clear enough. One hallucinated DROP COLUMN, one backfill with the wrong WHERE clause, and you're restoring from a backup at 2am.

Give it read-only access

…and copy-paste every statement yourself. Now you're the clipboard between your agent and your database, reviewing SQL in a terminal scroll with no diff, no history, no undo.

Code got pull requests twenty years ago. Your database is still running on trust.

How it works

Propose → Review → Apply

1

The agent proposes

Your agent connects to nuzur's MCP server. When it wants to add a table, alter a column, or backfill data, it doesn't execute anything — it creates a change request.

→
2

You review the diff

See exactly what will change: schema and data, side by side, before and after. Not a wall of SQL in a chat window — a structured diff against your actual model. Leave it in draft, edit it, or reject it.

→
3

Approve and apply

One click applies the change through nuzur — migration generated, executed, versioned. Full history of every change, who (or what) proposed it, and who approved it.

The part nobody else does: change requests cover data changes too, not just schema. "Backfill the category column for 40,000 rows" gets the same review-before-apply treatment as CREATE TABLE.

Setup

Connected in 5 minutes

  1. 1

    Sign up and import your database

    MySQL or PostgreSQL — cloud, or private networks via a local agent.

  2. 2

    Add the nuzur MCP server

    Add it to Claude Code, Cursor, VS Code, or any MCP client. One config block, hosted by us — nothing to run.

  3. 3

    Ask your agent to make a change

    It shows up in nuzur as a change request, waiting for you.

# add the nuzur MCP server
claude mcp add nuzur \
  https://ccmcp.nuzur.com
Get your MCP endpoint →
New · Agent-driven deploys

Your agent can ship it too

The same MCP server that keeps agents safe now lets them take a project all the way to production — with you approving every gate.

# in Claude Code, Cursor, or any MCP client
> /mcp__nuzur__deploy

Provider? digitalocean
API surface? both (rest + grpc)
→ running nuzur-cli deploy …
✓ live at https://api.yourapp.dev

One prompt, full deploy

A built-in MCP prompt hands your agent the entire deploy playbook: pick a provider, choose the API surface, run nuzur-cli deploy, and report back the live URL.

Granular tools, safe by design

Purpose-built MCP tools for schema edits, data change requests, queries, automations, and deployments — each scoped so the agent can only propose, never execute.

Your infrastructure, your credentials

Deploys land on your own Linux server, or a VM created through your own provider CLI — DigitalOcean, Hetzner, Linode, Google Cloud, Azure, Vultr, or Scaleway. nuzur never sees a cloud token.

nuzur is a full model-to-deploy platform — see the whole picture →
Why it works this way

The agent never holds the pen

nuzur is model-first: your database schema lives as a versioned model, and every change — human or agent — goes through it. That's what makes agent access safe by construction rather than by prompt engineering.

No direct write access

The agent's credentials can propose, not execute. There is no prompt injection that gets around a permission that doesn't exist.

Versioned everything

Every applied change request is a version you can inspect and trace. Your schema history reads like a git log.

Human-authored changes too

The same review flow works for your teammates. Agents just made it urgent.

Built in the open

Made by a solo founder, in public

FAQ

Questions, answered

Does the agent need my database credentials?

No. The agent talks to nuzur's MCP server. nuzur connects to your database — directly, or through a local agent for private networks. Your agent never sees a connection string.

What can the agent do without approval?

Read the model and query data (if you allow it), and create draft change requests. Nothing is applied to your database without a human approving it.

Which agents work with this?

Anything that speaks MCP: Claude Code, Claude Desktop, Cursor, VS Code, and any other MCP-capable client.

Which databases?

MySQL and PostgreSQL, including databases on private networks via nuzur local agents.

What does it cost?

Connecting an agent and reviewing change requests is free to start. See pricing for team plans.

Can I still edit the schema myself?

Yes — nuzur is a full visual database designer. The agent workflow sits on top of the same model, migrations, and versioning you'd use by hand.

Can my agent deploy the backend too?

Yes. The nuzur MCP server includes a deploy prompt that walks your agent through an entire production deploy with nuzur-cli — provisioning, database, HTTPS, and the live API. You approve the plan and the schema before anything runs.

Does nuzur get my cloud credentials?

No. Server provisioning runs through your own provider CLI (doctl, hcloud, gcloud, az, and more) already authenticated on your machine — or over SSH to a server you own. Credentials never pass through nuzur.

Give your agent a seatbelt

Stop choosing between "agent with prod access" and "human clipboard."

Connect your agent — free

Solo founder here — if setup takes you more than 5 minutes, reach out and I'll do it with you.

nuzur mascot