Kyvvu

Agent incident database

Supabase MCP: a support ticket got Cursor to leak secret tokens

6 Jul 2025Supabase · Supabase MCP server (used with Cursor)data exfiltrationdisclosed vulnerability

General Analysis showed that an attacker could write instructions into a customer support ticket that a developer's AI assistant later followed. The assistant read a table of secret tokens and posted them back into the ticket for the attacker to see.

How it unfolded

Each step on its own looked like ordinary agent work. The harm came from the sequence.

  1. 1

    Read a support ticket submitted by an outside attacker

  2. 2

    Model followed the instructions hidden in the ticket

  3. 3

    Queried the integration_tokens table with a role that bypasses row-level security

  4. 4

    Inserted the tokens as a new message in the attacker's ticket thread

    harm

Which policy would have stopped this?

In our reading of the reports, these policies from the open library stop the harmful step before it runs.

Open External content taint: high-impact actions require a fresh gate in the playground and build the path yourself.

This is our reading of public reporting. We haven't run it against the original systems.

What happened

In General Analysis's demonstration, an attacker submitted a support ticket through a public form with hidden instructions addressed to the Cursor assistant. A developer using Cursor with the Supabase MCP server later asked to see the latest open ticket, and the assistant fetched its messages, including the attacker's text. Running under the service_role, which bypasses row-level security, the assistant issued one SQL query that read the integration_tokens table and another that inserted its contents as a new message in the same ticket thread. The attacker then simply refreshed the ticket to read the tokens. The demonstration used dummy data.

Outcome

Supabase's documentation recommends read-only, project-scoped access and manual review of tool calls, and says its server wraps SQL results with instructions discouraging the model from following embedded commands. General Analysis notes that read-only mode blocks the write step but not the underlying risk.

Sources

A summary of public reporting. The step-by-step timeline is our reading of those sources; where they differ, we say so above.

Your agents

Could this happen to yours?

For security teams

Request a demo

We walk through your agents' riskiest paths with you.

For developers

Build a path yourself

Recreate a sequence in the playground and see how policies treat it.