@supabase/mcp-server-supabase

Executes arbitrary SQL queries against a Supabase PostgreSQL database by establishing authenticated connections through the Supabase client SDK. The MCP server translates query requests into native PostgreSQL operations, handling connection pooling, authentication via API keys, and result serialization back to the client. Supports parameterized queries to prevent SQL injection and returns structured result sets with column metadata.

Retrieves comprehensive schema information from Supabase PostgreSQL tables including column definitions, data types, constraints, indexes, and relationships. The MCP server queries PostgreSQL system catalogs (information_schema) to extract metadata without requiring manual schema definitions. Results include column nullability, default values, foreign key relationships, and primary key information.

Establishes WebSocket subscriptions to Supabase Realtime to receive live notifications when database records are inserted, updated, or deleted. The MCP server manages subscription lifecycle, filters changes by table and optional WHERE conditions, and streams change events to connected clients. Leverages Supabase's built-in Realtime infrastructure which uses PostgreSQL LISTEN/NOTIFY under the hood.

Manages Supabase Auth operations including user creation, password resets, email verification, and session management. The MCP server interfaces with Supabase Auth API to handle authentication flows, token generation, and user metadata updates. Supports both email/password and OAuth provider authentication, with built-in handling of JWT tokens and refresh token rotation.

Manages file uploads, downloads, and deletions in Supabase Storage buckets through the MCP server. The MCP server translates file operation requests into Supabase Storage API calls, handling multipart uploads, signed URLs for secure access, and bucket-level access control. Supports both public and private buckets with configurable retention and access policies.

Performs semantic similarity searches using PostgreSQL pgvector extension integrated with Supabase. The MCP server accepts vector embeddings (typically from an LLM embedding model) and executes similarity queries using cosine distance, L2 distance, or inner product operators. Results are ranked by similarity score and can be filtered by additional SQL conditions for hybrid search.

Enforces Supabase Row-Level Security policies by evaluating RLS rules before executing queries. The MCP server applies RLS policies based on the authenticated user's claims (from JWT tokens) and filters query results accordingly. Policies are defined in PostgreSQL and automatically enforced at the database level, preventing unauthorized data access.

Executes PostgreSQL stored procedures and functions through the MCP server, enabling complex database logic encapsulation. The MCP server translates function calls into native PostgreSQL function invocations, handling parameter passing, return value serialization, and error handling. Supports functions with multiple parameters, return types (scalar, composite, or set-returning), and transaction semantics.