Agent tools
Local MCP server with read-only SQL, dangerous-function denial, result limits, masking, and JSONL audit.
AI startups on Supabase
Ship AI features without exposing raw production data.
Protect the path your feature uses: MCP for agent tools, the PostgreSQL proxy for backend and ORM traffic, or the REST filter for supabase-js and PostgREST.
AI on SupabaseProtect each routed Supabase data path
Observe firstBACKEND SQL / WIRE
AI TOOLS / MCP
SUPABASE-JS / REST
Safe BoundarySurface-matched policy
ROUTED
SUPABASE PGWhy this exists
One product does not mean one interception trick
AI features reach Supabase through different protocols. Safe Boundary provides a dedicated enforcement surface for each major data path while keeping masking rules and audit fields aligned.
Start in observe mode, test the real application, then enforce. Keep RLS and least-privileged roles in place, and remove direct credentials or URLs from callers that must not bypass the boundary.
What it does
Concrete controls, stated precisely.
Application SQL
PostgreSQL wire proxy for ORMs and clients, with upstream role preservation by default.
supabase-js reads
PostgREST filter that forwards JWT and RLS semantics, masks responses, and blocks inference operators.
Surface decision
Pick the boundary by the interface your feature actually uses.
| Decision | If the feature uses… | Start with… | Validate… |
|---|---|---|---|
| AI tool calls | Local MCP | Tool list, row/byte limits, local audit | |
| Backend SQL / ORM | PostgreSQL wire | Driver compatibility, role preservation, direct DSN closure | |
| supabase-js / HTTP | PostgREST filter | JWT/RLS, response masking, inference operators |
How it works
A visible enforcement sequence.
Map paths
List every production database URL, REST URL, and MCP tool the feature can reach.
Choose surfaces
Place Safe Boundary on each in-scope access path.
Observe
Run production-like traffic and refine masking rules.
Close bypasses
Enforce policy and restrict direct credentials and network routes.
Protected when
- ✓Routed AI and application read paths
- ✓Configured sensitive PostgreSQL columns
- ✓Unsafe SQL and selected inference patterns
Important boundaries
- •Safe Boundary does not secure unrelated Supabase products automatically
- •Direct service_role or database credentials remain bypasses unless removed
- •It is a technical control, not automatic proof of regulatory compliance
Continue the story
Related Safe Boundary pages
MCP server →
Give AI tools a bounded PostgreSQL query surface with local audit.
Supabase REST protection →
Mask PostgREST responses and stop filters from becoming inference oracles.
Architecture →
See how the three enforcement surfaces share policy and audit semantics.
Test the boundary on a real access path.
Start in observe mode and validate the policy against representative traffic before you enforce it. That gives the team evidence for each decision, along with explicit bypass controls for the cases that need a deliberate exception.