The Autario MCP server is a data application infrastructure designed to integrate your tools with thousands of public datasets into a single normalized layer, and to run data applications on top of it. Its core value lies in placing your own platform data (ad accounts, stores, web analytics, social profiles, CRM, financial tools) alongside public data catalogs (World Bank, FRED, Eurostat, OECD, WHO, IMF, SEC, etc.) as context, so that your AI assistant can read tables, make joins, and explain data changes.
Main capabilities:
- Query your own platform data and public data, not just public data.
- Provides 57 MCP tools, each marked as read or write, so clients can control permissions accordingly.
- Cross-dataset joins: metrics from different datasets are automatically joined through the shared autario_time + autario_entity shadow columns, with no need to set up relationships manually; for example, get_entity_data(USA, [gdp, unemployment, life_expectancy]) can return a wide table.
- Built-in statistical primitives: correlate, regression, find_drivers, lag_analysis, seasonality_decomposition, etc., with output including effect sizes, p-values, and plain-language explanations.
- Chart persistence: publish_chart writes a Plotly specification to autario.com, generating a permanent, embeddable URL such as autario.com/chart/{slug}.
- Every public value cites its original source, and a verify_value tool is provided for verification.
- LLM-agnostic: compatible with Claude, GPT, Gemini, local models, and any other MCP-capable client.
How to use:
- Remote: point any HTTP MCP client at https://autario.com/mcp，无需安装。
- Local: run via npx autario-mcp over stdio.
- Example stdio configuration for Claude Desktop, Cursor, Cline, etc.: edit ~/.config/claude/claude_desktop_config.json (Mac/Linux) or %APPDATA%\Claude\claude_desktop_config.json (Windows).
- The hosted endpoint also supports MCP prompts: analyze-dataset, create-chart, compare-countries.
- Enabling write tools (publishing charts, creating datasets) requires adding API credentials: add them in the stdio configuration, or send them to the HTTP endpoint as x-api-key / x-api-secret headers. Keys are obtained at autario.com/account.
- Connectors (AUTARIO_API_KEY required): hosted, auto-refreshing REST-API tables, configured by the account owner in the Autario UI; agents can list and refresh them, and do not handle credentials.
- Build and publish data applications (AUTARIO_API_KEY required): the path is create_app -> update_app_manifest -> upload_app_bundle (or write_app_artifact) -> get_app_preview_url -> publish_app, with no review queue. Each tool is a thin wrapper around the same REST endpoint.
- autario.js: the in-app data SDK, running in a locked sandbox with no network. It reads data, and each call runs with the role of the person viewing the app. Use create_app datasets: [ids] to declare user-owned tables.
Use cases:
- Combine advertising, store, analytics, social, CRM, and financial data with public economic data for analysis, for example viewing spend, revenue, and ROAS by channel, or evaluating a sales decline alongside inflation.
- Group connected accounts by brand and instance, so agents can query a single brand without mixing in data from other brands.
- Get current quotes for public companies, avoiding outdated training data.
- Build and publish your own data applications, with no review required.
Limitations:
- Write tools and connectors require an API key.
- autario.js runs in a locked sandbox with no network.
- The read/write permission of each tool is determined by its own annotation.
- The public data catalog includes the World Bank, FRED, Eurostat, OECD, IMF, ECB, WHO, the United States Census Bureau, and user-contributed datasets.
Environment variables, the full tool reference, and development documentation are available at autario.com/developer/docs.
