Atlas
Wire a YAML semantic layer over your warehouse so Claude Code or Cursor agents query governed metrics instead of ad-hoc SQL.
Overview
Atlas is a MCP server for the Build phase that exposes a YAML-authored analytics semantic layer so AI agents query governed metrics over your warehouse or the NovaMart demo.
What is this MCP server?
- Human-authored YAML semantic layer with metrics and dimensions agents can resolve consistently
- Stdio MCP via npx @useatlas/mcp (v0.0.2) with optional ATLAS_API_URL to inherit API connections and governance
- Supports postgres, mysql, clickhouse, snowflake, duckdb, bigquery, and salesforce via ATLAS_DATASOURCE_URL
- Falls back to bundled NovaMart SQLite demo when no datasource URL is set
- Optional ATLAS_PROVIDER only when you also run Atlas chat or scheduler—the MCP server does not call an LLM directly
- MCP package version 0.0.2 (@useatlas/mcp, npm, stdio)
- Seven production datasource families named in docs plus bundled NovaMart SQLite demo
- Default Atlas API URL http://localhost:3001 when ATLAS_API_URL is unset
Community signal: 1 GitHub stars.
What problem does it solve?
Agents writing one-off SQL against production schemas produce wrong joins, inconsistent metrics, and risky queries when you need trustworthy numbers in the IDE.
Who is it for?
Solo builders shipping SaaS or internal tools who want Claude or Cursor to pull analytics through a defined YAML model tied to postgres, snowflake, bigquery, or the bundled demo.
Skip if: Teams that only need static BI dashboards with no agent access, or anyone unwilling to author and maintain a semantic YAML layer.
What do I get? / Deliverables
After you register @useatlas/mcp, your agent resolves business metrics through your semantic layer and optional Atlas API governance instead of improvising database access.
- MCP-accessible semantic layer aligned with your YAML definitions and optional Atlas API
- Agent-ready queries over postgres, mysql, clickhouse, snowflake, duckdb, bigquery, salesforce, or demo SQLite
- Shared governance context when ATLAS_API_URL points at a configured Atlas API instance
Recommended MCP Servers
Journey fit
Atlas is infrastructure you add while connecting product data to agent workflows—the canonical shelf is Build because you define entities and metrics before agents consume them in shipped features. Integrations fits an MCP bridge that inherits Atlas API connections, semantic definitions, and governance rather than owning UI or campaign analytics alone.
How it compares
MCP semantic analytics server, not a single Cursor skill or a standalone charting app.
Common Questions / FAQ
Who is Atlas for?
Indie and solo builders who use AI coding agents and want warehouse queries grounded in human-written metric definitions rather than raw schema dumps.
When should I use Atlas?
Use it while integrating analytics into agent workflows—defining YAML semantics before agents answer revenue, funnel, or ops questions against real or demo data.
How do I add Atlas to my agent?
Add the MCP server with npx @useatlas/mcp over stdio, set ATLAS_DATASOURCE_URL or ATLAS_API_URL as needed, and register the server in Claude Code, Cursor, or another MCP-compatible client.