Multi-Tenant PostgreSQL MCP Server

# Multi-Tenant PostgreSQL MCP Server A comprehensive, read-only Model Context Protocol (MCP) server for PostgreSQL databases with advanced multi-tenant support, schema introspection, and query capabilities. ## Features - 🔒 **Read-only by design** - All queries run in read-only transactions for safety - 🏢 **Multi-tenant support** - Access tables and functions across different schemas - 🔍 **Advanced schema introspection** - Detailed table structures, constraints, indexes, and DDL - 🔧 **Function definitions** - Complete function metadata and source code access - 📊 **Flexible querying** - Execute SQL queries with optional schema context - 🌐 **Network-ready** - Connect to local or remote PostgreSQL instances ## Installation ### Via npm (recommended) ```bash npm install -g @ahmetkca/mcp-server-postgres ``` ### Via npx (no installation required) ### Run directly (no install) ```bash npx @ahmetkca/mcp-server-postgres "postgres://user:password@host:port/database" ``` ### From source ```bash git clone https://github.com/ahmetkca/mcp-server-postgres.git cd mcp-server-postgres npm install npm run build ``` ## Usage ### Command Line ### With global installation: ```bash mcp-server-postgres "postgres://user:password@host:port/database" ``` ### With npx (no installation required) ### Direct execution (npx) ```bash npx @ahmetkca/mcp-server-postgres "postgres://user:password@host:port/database" ``` ### With Claude Desktop Add to your Claude Desktop MCP configuration (`~/Library/Application Support/Claude/claude_desktop_config.json` on macOS): ### Option 1: Using npx (recommended - no installation required): ```json { "mcpServers": { "postgres-multitenant": { "command": "npx", "args": ["-y", "@ahmetkca/mcp-server-postgres", "postgres://user:password@localhost:5432/mydb"] } } } ``` ### Option 2: Using global installation ### Global install ```json { "mcpServers": { "postgres-multitenant": { "command": "mcp-server-postgres", "args": ["postgres://user:password@localhost:5432/mydb"] } } } ``` ### With other MCP clients The server uses stdio transport, so it can be used with any MCP client that supports subprocess communication: ```bash # Direct execution npx @ahmetkca/mcp-server-postgres "postgres://connection-string" # Or with global install mcp-server-postgres "postgres://connection-string" ``` ## Connection String Examples ### Local PostgreSQL ```bash mcp-server-postgres "postgres://user:password@localhost:5432/mydb" ``` ### Remote PostgreSQL ```bash mcp-server-postgres "postgres://user:password@db.example.com:5432/mydb" ``` ### AWS RDS ```bash mcp-server-postgres "postgres://user:password@mydb.abc123.us-east-1.rds.amazonaws.com:5432/mydb" ``` ### Google Cloud SQL ```bash mcp-server-postgres "postgres://user:password@1.2.3.4:5432/mydb" ``` ### With SSL (recommended for production) ```bash mcp-server-postgres "postgres://user:password@host:5432/mydb?sslmode=require" ``` ## Available Resources ### Schema Overview - **URI Pattern**: `pg-schema://{schemaName}/overview` - **Description**: High-level overview of a schema including counts (tables, views, functions) and quick links (URIs) to table structures and function definitions for drill-down - **Example**: `pg-schema://public/overview` - **Completions**: Supports typeahead for `schemaName` ### Table Structures - **URI Pattern**: `pg-table://{schemaName}/{tableName}/structure` - **Description**: Comprehensive table metadata including columns, constraints, indexes, and DDL - **Example**: `pg-table://public/users/structure` ### Function Definitions - **URI Pattern**: `pg-func://{schemaName}/{functionName}/(identityArgs)/definition` - **Description**: Complete function metadata, parameters, return types, and source code - **Example**: `pg-func://public/calculate_total/()/definition` - Overloads: identityArgs is the canonical PostgreSQL identity argument list (e.g. `(integer,integer)`). Example overload: `pg-func://public/add/(integer,integer)/definition` ## Available Tools ### 1. Multi-Tenant Database Query Tool Execute read-only SQL queries to retrieve, analyze, and explore PostgreSQL data across multiple tenant schemas. Supports complex SQL including JOINs, CTEs, window functions, and aggregations. All queries run in read-only transactions for safety. **Parameters:** - `sql` (string, required): The SQL query to execute. Can be any valid PostgreSQL SELECT statement including complex queries with JOINs, CTEs, window functions, aggregations, etc. Will be executed in a read-only transaction for safety. - `schema` (string, optional): Optional schema name (tenant) to set as search_path before executing the query. When specified, unqualified table names will resolve to tables in this schema. Use this for tenant-specific queries. - `explain` (boolean, optional): Set to true to return the query execution plan instead of query results. Useful for performance analysis and optimization. Returns PostgreSQL EXPLAIN output in JSON format. **Examples:** ```sql -- Simple query with default schema SELECT * FROM users LIMIT 10; -- Complex query with JOINs and aggregations SELECT u.name, COUNT(o.id) as order_count, SUM(o.total) as total_spent, AVG(o.total) as avg_order_value FROM users u LEFT JOIN orders o ON u.id = o.user_id WHERE u.created_at > '2024-01-01' GROUP BY u.id, u.name ORDER BY total_spent DESC; -- Query with CTE and window functions WITH monthly_sales AS ( SELECT DATE_TRUNC('month', created_at) as month, SUM(total) as monthly_total, LAG(SUM(total)) OVER (ORDER BY DATE_TRUNC('month', created_at)) as prev_month FROM orders GROUP BY DATE_TRUNC('month', created_at) ) SELECT month, monthly_total, ROUND(((monthly_total - prev_month) / prev_month * 100)::numeric, 2) as growth_rate FROM monthly_sales WHERE prev_month IS NOT NULL; -- Tenant-specific query (set schema parameter to "tenant_a") SELECT * FROM orders WHERE status = 'pending'; -- Performance analysis (set explain parameter to true) SELECT u.*, o.total FROM users u JOIN orders o ON u.id = o.user_id; ``` ### 2. List Database Schemas Discover all available database schemas (tenants) with statistics including table and function counts. Use this to explore the multi-tenant structure, identify available tenants, or get an overview of database organization. **Parameters:** - `include_system` (boolean, optional): Set to true to include PostgreSQL system schemas (information_schema, pg_catalog, pg_toast) in the results. Default false shows only user/tenant schemas. Use true for administrative or debugging purposes. **Use Cases:** - Explore multi-tenant database structure - Identify available tenant schemas - Get overview of database organization - Administrative schema analysis ### 3. Describe Schema Get comprehensive information about a specific database schema (tenant) including detailed statistics, all tables, views, functions, and custom types. Use this to understand a tenant's database structure, analyze schema composition, or prepare for schema-specific operations. **Parameters:** - `schema_name` (string, required): Name of the database schema (tenant) to analyze. Must be an exact schema name from the database. Use list-schemas tool first to discover available schema names. **Returns:** - Schema statistics (table count, view count, function count, type count) - Detailed table information with column counts - Function definitions and metadata - Organized metadata perfect for schema analysis and documentation ## Multi-Tenant Architecture This server is designed for multi-tenant PostgreSQL setups where: - Different tenants have separate schemas (e.g., `tenant_a`, `tenant_b`, `public`) - Each schema contains tenant-specific tables and functions - You need to query across different tenant contexts ### Example Multi-Tenant Usage ```sql -- List all schemas -- Use "list-schemas" tool -- Describe a specific tenant schema -- Use "describe-schema" tool with schema_name: "tenant_a" -- Query tenant-specific data -- Use "query" tool with schema: "tenant_a" SELECT * FROM orders WHERE status = 'pending'; -- Get table structure for a tenant -- Access resource: pg-table://tenant_a/orders/structure ``` ## Security Features - **Read-only transactions**: All queries are wrapped in `BEGIN TRANSACTION READ ONLY` - **No data modification**: INSERT, UPDATE, DELETE, and DDL operations are blocked - **Schema isolation**: Optional schema context prevents cross-tenant data access - **Connection pooling**: Efficient resource management with automatic cleanup ## Development ### Prerequisites - Node.js 18+ - PostgreSQL database (local or remote) - TypeScript knowledge (for contributions) ### Setup ```bash git clone https://github.com/ahmetkca/mcp-server-postgres.git cd mcp-server-postgres npm install ``` ### Development Commands ```bash npm run dev # Watch mode for development npm run build # Build for production npm run start # Start the built server ``` ### Testing ### Start the server with a test database ```bash npm run build npm start "postgres://user:password@localhost:5432/testdb" ``` ### In another terminal, test with MCP Inspector ```bash npx @modelcontextprotocol/inspector ``` ## Troubleshooting ### Connection Issues - Verify your PostgreSQL connection string - Check network connectivity to the database - Ensure the database user has SELECT permissions - For remote connections, verify firewall settings ### Permission Issues - The database user needs SELECT permissions on: - All tables you want to query - `information_schema` views - `pg_catalog` system tables (for function definitions) ### Schema Access - Ensure the database user has USAGE permission on schemas - For multi-tenant setups, grant access to relevant tenant schemas ## Contributing 1. Fork the repository 2. Create a feature branch (`git checkout -b feature/amazing-feature`) 3. Commit your changes (`git commit -m 'Add amazing feature'`) 4. Push to the branch (`git push origin feature/amazing-feature`) 5. Open a Pull Request ## Changelog ### v1.1.0 - 2025-08-07 - Adopt custom resource URI schemes: - Tables: `pg-table://{schemaName}/{tableName}/structure` - Functions: `pg-func://{schemaName}/{functionName}/(identityArgs)/definition` - Per-overload function resources using canonical `identityArgs` to disambiguate overloads - Deterministic, deduplicated resource listings using `DISTINCT ON` and stable ordering - Exact overload resolution via `regprocedure`/OID for function metadata - README updated to reflect new URI schemes and best practices - TypeScript build fixes: NodeNext ESM config, top-level await, and `pg` Pool named import ### v1.0.0 - Initial release - Multi-tenant PostgreSQL support - Advanced schema introspection - Function definition access - Read-only query execution - Comprehensive documentation ## Examples: listing and reading Below are minimal examples of how resources appear in clients and how to read them. Actual UX varies by MCP client, but the URIs and fields are the same. #### 1) Listing resources (conceptual) Clients request a list from the server and render entries like: ``` name: public uri: pg-schema://public/overview — name: public.users uri: pg-table://public/users/structure — name: tenant_a.app_event uri: pg-table://tenant_a/app_event/structure — name: public.add(integer,integer) uri: pg-func://public/add/(integer,integer)/definition — name: public.add(text,text) uri: pg-func://public/add/(text,text)/definition ``` #### 2) Reading a schema overview - Select: `pg-schema://public/overview` - The client sends a read request and receives JSON with statistics and quick links (URIs) to tables and functions. JSON-RPC (abridged) example over stdio: ```json { "jsonrpc": "2.0", "id": 1, "method": "resources/read", "params": { "uri": "pg-schema://public/overview" } } ``` Response (shape): ```json { "jsonrpc": "2.0", "id": 1, "result": { "contents": [ { "uri": "pg-schema://public/overview", "mimeType": "application/json", "text": "{ ... statistics, tables.first_100 [{name, uri}], functions.first_100 [{name, uri}] ... }" } ] } } ``` #### 2) Reading a table structure - Select: `pg-table://tenant_a/app_event/structure` - Client sends a read request and receives JSON describing columns, constraints, indexes, relationships, and DDL. JSON-RPC (abridged) example over stdio: ```json { "jsonrpc": "2.0", "id": 1, "method": "resources/read", "params": { "uri": "pg-table://tenant_a/app_event/structure" } } ``` Response (shape): ```json { "jsonrpc": "2.0", "id": 1, "result": { "contents": [ { "uri": "pg-table://tenant_a/app_event/structure", "mimeType": "application/json", "text": "{ ... columns, constraints, indexes, ddl ... }" } ] } } ``` #### 3) Reading a specific function overload - Select: `pg-func://public/add/(integer,integer)/definition` - The server resolves the exact overload by `regprocedure` and returns the definition. JSON-RPC (abridged) example: ```json { "jsonrpc": "2.0", "id": 2, "method": "resources/read", "params": { "uri": "pg-func://public/add/(integer,integer)/definition" } } ``` Response (shape): ```json { "jsonrpc": "2.0", "id": 2, "result": { "contents": [ { "uri": "pg-func://public/add/(integer,integer)/definition", "mimeType": "application/json", "text": "{ ... parameters, return_type, language, volatility, security, source, ddl ... }" } ] } } ``` Tips - Choose the exact overload by picking the URI that includes the identity args in parentheses. - All identifiers in URIs are URL-encoded; clients decode them before invoking the handler. - Lists are deduplicated and ordered for a clean browsing experience. ## License This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.