MCP 服务器

RunContext

Enables AI agents to understand and query your database safely by providing a semantic layer of metadata, with tools to search, explain, validate, and generate safe SQL.

README

<h1 align="center">RunContext</h1>

<p align="center"> <strong>Turn your database into an AI-ready semantic plane.</strong> </p>

<p align="center"> Run <code>context setup</code>, connect your database, and RunContext introspects the schema to scaffold a semantic plane — structured metadata that tells AI agents what your data means, how to query it safely, and which business rules apply. Then hand off to your AI agent to curate it to Gold. Serve it via MCP to Claude Code, Cursor, Copilot, or any compatible tool. </p>

Quick Start

npx @runcontext/cli setup

A browser wizard opens. Connect your database, fill out the Context Brief, and RunContext scaffolds your semantic plane automatically.

How It Works

The context setup wizard walks you through six steps:

Step	What happens
1. Connect	Connect to your database
2. Define	Name your semantic plane, set owner and sensitivity
3. Scaffold	Auto-introspect schema and generate Bronze-tier metadata
4. Checkpoint	Review what was built
5. Curate	Hand off to your IDE's AI agent to curate metadata to Gold using real data queries
6. Serve	Start the MCP server and serve your semantic plane to AI tools

The result is a context/ directory of YAML files following the OSI (Open Semantic Interchange) specification.

Tiers

Bronze -- Scaffolded automatically. Every model and field has a description, owner, and type. Your data is discoverable.

Silver -- Sample values, tags, and richer descriptions. AI agents can interpret the data with confidence.

Gold -- Golden queries, guardrails, semantic roles, and business rules. Agents generate correct SQL on the first try.

Check your progress with context tier.

Commands

context setup                  # Browser wizard -- database to semantic plane
context tier [model]           # Bronze/Silver/Gold scorecard
context serve                  # Start MCP server (stdio for IDE integration)
context lint                   # Validate metadata
context fix                    # Auto-fix lint issues
context verify                 # Check metadata against live database
context build                  # Compile context layer to manifest
context dev                    # Watch mode
context introspect             # Schema introspection
context explain [term]         # Look up models, terms, or owners

MCP Server

Add RunContext to your AI tool's MCP configuration:

{
  "mcpServers": {
    "runcontext": {
      "command": "npx",
      "args": ["@runcontext/cli", "serve", "--stdio"]
    }
  }
}

6 MCP tools available:

Tool	Description
`context_search`	Find models, datasets, and fields by keyword
`context_explain`	Get full model details
`context_validate`	Run the linter
`context_tier`	Get tier scorecard
`context_golden_query`	Find validated SQL queries
`context_guardrails`	Get WHERE clauses and access rules for tables

Works with Claude Code, Cursor, Copilot, Windsurf, Claude Desktop, and any MCP-compatible tool.

Database Support

Database	Supported
PostgreSQL	Yes
DuckDB	Yes
MySQL	Yes
SQL Server	Yes
SQLite	Yes
Snowflake	Yes
BigQuery	Yes
ClickHouse	Yes
Databricks	Yes

Packages

Package	Description
`@runcontext/cli`	CLI, setup wizard, and MCP server
`@runcontext/core`	Compiler, linter, tier engine, adapters
`@runcontext/mcp`	MCP protocol implementation
`@runcontext/ui`	Browser-based setup wizard and planes viewer
`@runcontext/uxd`	Design system components
`@runcontext/db`	Database utilities
`create-runcontext`	Project scaffolder (`npx create-runcontext`)