Allows deployment of the MCP server on Cloudflare Workers, providing a serverless runtime for the MCP implementation.
Enables direct SQL execution against PostgreSQL databases with connection pooling, safe query generation, and parameterized queries. Supports searching, retrieving, and managing data in Postgres databases through SQL statements defined in YAML configuration.
@vibase/core
Easily read and mutate Postgres data using MCP.
Overview
@vibase/core
provides the foundational components for creating MCP (Model Context Protocol) servers on top of Postgres using simple configuration files.
- Run on your own machine with STDIO
- Deploy on lambda or Cloudflare workers
- Package in a Claude Desktop Extension
Features
- YAML Configuration: Define MCP tools using simple YAML configuration files
- PostgreSQL Support: Direct SQL execution against PostgreSQL databases with connection pooling
- Safe SQL Generation: Uses parameterized queries for safe SQL query construction
- Type Safety: Full TypeScript support with Zod validation
- Connection Management: Automatic connection pooling and cleanup
Installation
Quick Start
- Create a YAML configuration file (
tools.yaml
):
- Create and run the MCP server:
Your MCP server now runs over streamable HTTP on http://localhost:5555
You can view options with --help
API Reference
Use Vibase with any TypeScript MCP server.
Configuration
loadConfigFromYaml(configPath: string): ToolboxConfig
Loads and validates configuration from a YAML file.
Parameters:
configPath
: Path to the YAML configuration file
Returns:
ToolboxConfig
: Validated configuration object
Throws:
- Configuration validation errors
- File not found errors
- YAML parsing errors
loadConfigFromYamlString(yamlContent: string): ToolboxConfig
Loads and validates configuration from a YAML string.
Parameters:
yamlContent
: YAML configuration as a string
Returns:
ToolboxConfig
: Validated configuration object
Throws:
- Configuration validation errors
- YAML parsing errors
Server Creation
createMcpServerFromConfig(config: ToolboxConfig, options?: ServerOptions): { server: McpServer; cleanup: () => Promise<void>; plugins: PluginRegistry }
Creates an MCP server from a validated configuration object.
Parameters:
config
: Validated configuration objectoptions
: Optional server configuration options
Returns:
server
: The MCP server instancecleanup
: Async function to clean up database connectionsplugins
: Plugin registry for registering custom plugins
addToolsToMcpServer(server: McpServer, config: ToolboxConfig): { cleanup: () => Promise<void>; plugins: PluginRegistry }
Adds tools from configuration to an existing MCP server. This is useful when you want to add tools to a server that was created separately.
Parameters:
server
: Existing MCP server instanceconfig
: Validated configuration object
Returns:
cleanup
: Async function to clean up database connectionsplugins
: Plugin registry for registering custom plugins
Example:
Configuration Format
Sources
Define database connections:
Tools
Define SQL-based tools:
Plugins API
@vibase/core
includes a plugin system that allows you to modify query execution behavior. The most common use case is implementing authentication with JWT tokens and PostgreSQL Row Level Security (RLS).
Plugin Registration
JWT Authentication Plugin Example
Here's how you can implement JWT bearer token authentication:
How Plugins Work
- Plugin Structure: Plugins are objects with a
name
andcallbacks
for different hooks - Plugin Registration: Use
plugins.register(plugin)
to register a complete plugin - Hook Execution: Plugin callbacks run before each query (more hooks coming soon)
- Context Access: Callbacks receive context including tool name, parsed arguments, and request metadata
- Session Variables: Plugins return
pgSettings
Maps that become PostgreSQL session variables - Automatic Transactions: When
pgSettings
are present, queries run in transactions with variables applied - RLS Integration: SQL queries can access variables via
current_setting('jwt.claims.sub')
Plugin Management
Row Level Security Integration
With the authentication plugin, your SQL queries can reference JWT claims:
Plugin Context
Plugin callbacks receive a context object with:
toolName
: Name of the tool being executedtoolConfig
: Tool configuration from YAMLpool
: Database connection poolparsedArgs
: Validated tool argumentsextra
: MCP request metadata (authInfo, requestInfo, etc.)query
: The SQL query being executed
Security Considerations
- Always use parameterized queries (enforced by this package)
- Store database credentials securely (consider environment variables)
- Limit database user permissions to only what's needed
- Use SSL connections for production databases
- Validate and sanitize user input before using in queries
Contributing
Contributions are welcome!
License
Apache 2.0
This server cannot be installed
Easily query and mutate Postgres data using MCP. This is an open source project.
Related MCP Servers
- -securityAlicense-qualityAn MCP server that connects to Supabase PostgreSQL databases, exposing table schemas as resources and providing tools for data analysis through SQL queries.Last updated -JavaScriptMIT License
- -securityAlicense-qualityA template project for building custom MCP servers that enables direct access to PostgreSQL databases, allowing SQL query execution and schema information retrieval through the Model Context Protocol.Last updated -3PythonMIT License
- -securityFlicense-qualityPG-MCP is an HTTP server implementation that enables AI systems to interact with PostgreSQL databases via MCP, providing tools for querying, connecting to multiple databases, and exploring schema resources. The system enriches context by extracting table/column description from database catalogs.Last updated -8
- AsecurityAlicenseAqualityA blazing fast MCP server that enables AI agents to interact with multiple PostgreSQL databases, providing functionality to list tables, inspect schemas, execute queries, and run transactions.Last updated -458TypeScriptMIT License