MCP overview

Astrato’s Model Context Protocol (MCP) server lets an AI assistant work with your semantic layers. It can list them, search for fields, answer business questions that are already defined, and make and publish changes when you ask. The assistant acts with your Astrato account and permissions.

Availability: MCP must be enabled for your Astrato environment. Ask your Astrato contact to confirm access before configuring a client.

Endpoint

• MCP URL: https://app.astrato.io/mcp
• Transport: remote Streamable HTTP
• Authentication: OAuth for interactive clients, or an Astrato Personal Access Token (PAT)

Examples use https://app.astrato.io. If your organisation provides a private or custom deployment origin, replace the Astrato origin in the URLs.

Connecting AI assistants

Choose your assistant. Each tile opens its setup steps in Connect an AI assistant.

ClaudeClaude.ai and the desktop app. Claude CodeThe command-line and IDE coding agent. Snowflake Cortex CodeSnowflake's coding assistant, also called CoCo. ChatGPTAdd Astrato as a plugin in ChatGPT. Visual Studio CodeAdd Astrato from the Chat panel. OpenAI CodexCreate and install an Astrato plugin.

Technical guides

Configure an MCP clientDeveloper settings, OAuth client IDs and personal access tokens. Scopes and permissionsWhat each scope allows, and how your role limits it. Query governed dataChoose between stored questions, governed items and raw fields. Results and errorsResult shapes, error codes, response budgets and query replays. Data connections and importsCreate and browse connections, and import CSV files, without sharing credentials in chat. Tool referenceEvery tool, the scopes it needs and what it does. Administration and limitsTenant user updates, exit-node routing and what MCP does not cover.

Example tools

These four tools cover the usual path from a new semantic layer to a published Data App. The tool reference lists every tool.

• astrato_semantic_layer_create creates an empty, unpublished semantic layer on a datasource.
• astrato_semantic_layer_apply_measure_change adds or changes a measure in the layer’s draft.
• astrato_semantic_layer_publish publishes the layer, so queries and Data Apps can use it.
• astrato_dashboard_freeform_create reserves an unpublished Data App. Publish its content with the CLI.

Skills

Astrato’s agent skills are portable instruction folders that teach an assistant Astrato workflows. They are published at github.com/astrato-io/astrato-skills. Install the astrato skill as the entry point, then add the task skills you need. See Astrato skills for what each one covers.

Capabilities

The tools use one product-shaped namespace:

• astrato_data_* manages connections, sources, and governed data queries.
• astrato_workspace_* discovers and organizes visible Astrato resources.
• astrato_semantic_layer_* discovers and authors semantic models and their business questions.
• astrato_dashboard_canvas_* manages Canvas dashboard lifecycle operations.
• astrato_dashboard_freeform_* creates unpublished Data Apps and manages their metadata, hosting and lifecycle. Publish content through the CLI.
• astrato_management_access_* manages direct grants over typed Astrato resources.
• astrato_get_context reports the live tool contract, capabilities, permissions, authenticated Astrato identity, and routing context.

OAuth scopes control authority independently of these product capabilities.

The identity context includes the current user’s name, email, and tenant role from verified Astrato identity data. These values explain the session but never grant access; every operation is authorized again by its scope and owning Astrato service. Group memberships are not injected.

The gateway may expose fewer tools when a product capability is disabled for your environment. Always use live tool discovery rather than assuming that every documented tool is available.

Nash is the agent identity. Capabilities are the Astrato product operations Nash can use, while skills are optional workflow guidance for combining those operations; neither a capability nor a skill is a separate Nash agent.

The astrato_management_* area starts with access control. Tenant settings, the Astrato user and group directory, licensing, identity policy, integrations, embedding, AI governance, tenant assets, and exports join it only when executable, server-authorized tools exist. Reporting remains a separate capability because schedules and subscriptions are not uniformly administrator-only. Unimplemented roadmap capabilities are never advertised by MCP discovery.