Provides agent-safe control plane for Supabase, enabling AI agents to manage schema, edge functions, auth workflows, and storage with plan/approve workflows and audit trail.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@diabase plan to create a new 'analytics' table in Supabase"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Read this in: Italiano
Agent-safe control plane for Supabase, self-hosted or cloud.
Diabase lets AI agents (Claude Code, or any LLM via API) operate and manage your entire Supabase backend — schema, edge functions, auth workflows, storage — while you watch every move from a modern GUI and stay in charge of the decisions that matter.
⚠️ Status: early development. Nothing here is ready for production use yet. Follow along or star the repo — the first usable release will be announced.
Supabase's official MCP executes directly against your production project: no plan, no diff, no staged approval, no audit trail. Diabase is the opinionated layer that makes agent-driven backend management safe enough for a company to allow:
Plan & Approve — large operations become a readable plan you approve, reject, or amend; destructive steps are highlighted
Audit trail — every action, human or AI, recorded append-only
Adapters — the same experience against Supabase self-hosted or Supabase Cloud
Projects — Claude-style workspaces with their own system prompt, context files and autonomy level
Your LLM, your choice — Claude Code CLI (subscription) or pay-per-use API, configured in settings
Related MCP server: Supabase MCP
docker compose up -dOpen http://localhost:8000 and sign in with admin / admin — you are forced
to rotate the password on first login. Data (database and generated secret key)
lives in the diabase-data volume; set DIABASE_PORT or DIABASE_ALLOWED_HOSTS
in the environment to change the defaults.
Optional knowledge graph — a temporal graph memory (Graphiti + Neo4j) over audited actions and chat:
docker compose --profile graph up -dThen enable it in the graph settings with two connections: an extraction LLM (Anthropic or OpenAI-compatible) and an OpenAI-compatible embedder (Ollama works). Without the profile — or with Neo4j down — Diabase behaves exactly as before: the graph is an enhancement, never a requirement.
The Claude Code CLI backend needs the CLI installed, which the image does not ship. Inside the container use API connections (Anthropic or any OpenAI-compatible endpoint); the Claude Code backend works in a source checkout.
For development, run from source instead:
uv run python manage.py migrate
uv run python manage.py runserver 8001This server cannot be deployed
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
MCP server for building and testing AI agents with multi-model experimentation and insights.
MCP server connecting AI agents to 100+ apps (Gmail, Slack, Notion, GitHub) via one-click OAuth.
Zero-secret MCP gateway for AI agents: risk-scored, audited calls with human-in-the-loop approval.
