The Supabase MCP server lets Claude, Cursor, ChatGPT, and AI agents read and act inside Supabase. Here's the official setup, the real security risks, and how to deploy it with DLP-grade redaction at the MCP layer.
The Supabase MCP server is a Model Context Protocol implementation that exposes Supabase's API as a standardized set of tools to AI agents. Once connected, an agent like Claude can perform run SQL queries, list tables, read rows, inspect schemas, and execute migrations on the authenticated user's behalf — turning Supabase's API surface into AI-actionable capabilities.
Refer to the official Supabase MCP server documentation for the current tool list, OAuth scopes, and rate-limit behavior. The setup pattern is consistent with other MCP integrations: an OAuth client ID/secret, a custom connector in Claude (or another MCP-aware AI client), and the server starts serving tool calls.
From the user's perspective, the AI agent suddenly knows their Supabase. From the security perspective, the AI agent now has read access — and often write access — to every record the user can touch in Supabase.
That's the value. It's also where security teams need a control layer.
The risks fall into four categories that every healthcare, fintech, and enterprise security team should price into the deployment.
1. SQL tool calls return raw rows in bulk. execute_sql runs arbitrary SELECTs. A single query against a users or orders table returns emails, addresses, payment references, and PII for the whole table — no row limit, no field masking.
2. The auth schema is one query away. Supabase’s auth.users holds every registered user’s email, phone, and provider identity. MCP tool calls can read it directly, handing the model your entire user base.
3. Schema introspection maps where sensitive data lives. list_tables and column introspection hand the agent — and the model — a full map of the database, accelerating both legitimate use and accidental exfiltration.
4. Write and migration tools mutate production. execute_sql and migration tools can INSERT, UPDATE, DELETE, and alter schema. A confused or compromised agent can modify or destroy regulated data, not just read it.
The traditional DLP a company already runs — at the network edge, on the file share, inside the SaaS-native rule engine — does not sit in the MCP path. The tool response goes straight from Supabase into the AI agent's context window. That gap is where Strac Supabase MCP DLP lives.
Strac's Supabase MCP DLP sits between AI agents and the Supabase MCP server. Every tool call passes through Strac's MCP-layer inspection before content reaches the AI agent's context window. Sensitive content is redacted, tokenized, pseudonymized, or vaulted depending on policy. Non-sensitive content flows through untouched.
What this looks like in practice:
The same Strac MCP DLP layer covers Claude Cowork, Slack MCP, and other surfaces — one control plane across every place AI agents touch your regulated data.
The screenshot below shows Strac's MCP DLP redacting sensitive data from a real Claude session — patient identifiers, customer emails, and credit card numbers tokenized inline before the model received the prompt. The same inspection pattern runs on every Supabase MCP tool call routed through Strac.
Setup is agentless and takes under 10 minutes.
.webp)
json
"mcpServers": {
"supabase": {
"url": "https://mcp.strac.io/supabase",
"auth": { "type": "bearer", "token": "<your-strac-token>" }
}
}
For Cursor, OpenAI Agents, custom agents — same endpoint, same auth.The same Strac Supabase MCP DLP control produces evidence mapped to every major compliance framework.
For the broader AI-data-governance program this sits inside, see the AI Data Governance framework.
The Supabase MCP server is a Model Context Protocol implementation that lets AI agents (Claude, Cursor, ChatGPT, Perplexity, custom agents) read and act inside Supabase via standardized tool calls. It's how an AI assistant gets contextual access to every table, row, column, and schema in the connected Postgres database — including auth tables, user profiles, and application data.
By itself, no — not without an additional DLP layer. The Supabase MCP server honors the authorizing user's permissions but returns whatever that user can see, including PII, PHI, credentials, source code, and other regulated content. For enterprise use with regulated data, you need an MCP-layer DLP control like Strac Supabase MCP DLP that inspects and redacts every tool response before content reaches the AI model.
Supabase's built-in protections operate at the storage and policy layer — sensitivity labels, retention policies, native DLP rules at posting/sharing time. None of those sit in the MCP tool-call path by default. Strac is purpose-built for the MCP layer: it inspects every tool response before content reaches the AI agent's context window, with detection breadth (PII / PHI / PCI / secrets / source code / OCR-in-images) that goes well beyond most native rule engines.
Yes. Strac exposes a standard MCP endpoint, so any MCP-aware AI client routes tool calls through it with one configuration change. No SDK changes, no application code changes.
PII (SSN, driver's license, passport, address, phone, email), PHI (clinical notes, MRN co-occurrence, ICD-10 codes adjacent to identifiers, lab values), PCI (full and partial card numbers via Luhn check), credentials (API keys, AWS / GCP / Azure access keys, OAuth tokens, JWTs, SSH keys, private keys — 48+ patterns), proprietary content (M&A keywords, source code fingerprints), and custom detectors trained on your internal data classifications. Detection runs across text, files, images (OCR), and structured fields.
Under 10 minutes for the first workspace. OAuth Strac into Supabase, paste the Strac MCP endpoint into your AI client's config, pick a policy template, done. No agents to install, no Supabase re-permissioning, no application code changes.
Redacted content is replaced inline in the tool response. Optionally, sensitive content can be vaulted — replaced with a short-lived retrieval link that only authorized users can resolve, so the original data is retrievable for legitimate use without ever entering the AI context. Vaulted data is stored encrypted at rest in your Strac tenant; you control retention.
Yes. Strac produces a per-call audit log: timestamp, AI client identity, user, tool invoked, resource accessed, data classes detected, redactions applied, vault references, disposition. The log is queryable in the Strac console and exportable to your SIEM. This is the evidence trail SOC 2, HIPAA, PCI, and GDPR auditors will ask about for AI-agent activity in Supabase.
The Supabase MCP server is rapidly becoming the way AI agents read into Supabase. That surface contains every category of regulated and proprietary data your organization has. Running Supabase MCP in 2026 without an MCP-layer DLP control is not a question of if the first incident reaches your security team; it's when.
Strac Supabase MCP DLP gives you the protection layer, the audit evidence, and the framework-agnostic compliance coverage so you can let your team use Supabase with Claude, Cursor, Cowork, ChatGPT, and any future AI client without making each one a separate security exception.
If you are running — or about to run — Supabase MCP in production, book a 30-minute demo. We'll walk through the architecture, the policy templates, and a deployment plan for your specific Supabase workspace and AI clients.
For the broader MCP DLP control plane across every SaaS surface, see the MCP DLP pillar. For more SaaS-specific deep dives: Slack MCP, Google Workspace MCP, Gmail MCP, Google Drive MCP, Microsoft 365 MCP, Notion MCP, Jira MCP.
.avif){kind=icon}
.gif)
