Orion by Gravity MCP Server

Remote MCP server for Orion, Gravity's autonomous AI data analyst. Query metrics, run analyses, fetch dashboards, and surface anomalies from your data stack.

Data & Enrichment by Gravity OAuth2 active

Connect (Remote Server) Docs

Overview

Orion by Gravity is an autonomous AI analyst that connects to data sources like Looker, BigQuery, Snowflake, Databricks, and dbt, then proactively investigates metrics, surfaces anomalies, and produces narrative insights, dashboards, and slide decks. The Orion MCP server is a remote, provider-hosted endpoint that lets Claude, Cursor, and other MCP-compatible AI assistants interact with Orion projects, metrics, workflows, and knowledge bases without leaving the chat.

The server exposes 20 tools split into 5 action tools (asking Orion questions, creating conversations, executing metrics, running workflows, deleting artifacts) and 15 read-only tools (listing projects, users, conversations, metrics, workflows, artifacts, recommendations, plus retrieving metric results, insights, and wiki content). Long-running analyses through ask_orion are handled with built-in polling, so agents can wait for multi-step investigations to complete.

The server is hosted at https://g.runorion.com/mcp over HTTP transport. There is no local installation: users authenticate via a browser sign-in with their Orion account, and the connector works out of the box in Claude Desktop's connector marketplace, Claude Code, and Cursor.

Tools

Tool	Description
`ask_orion`	Send a natural language question to Orion and get an AI-powered data analysis response. Handles multi-step analyses with automatic polling.
`create_conversation`	Start a new Orion chat conversation.
`execute_metric`	Trigger execution of a metric to refresh its current value.
`run_workflow`	Execute an insight maker workflow.
`delete_artifact`	Remove a slide deck, dashboard, or report.
`list_projects`	List Orion projects accessible to the user.
`list_users`	List all organization users with IDs, names, and roles.
`list_conversations`	List a user's chat conversations.
`get_conversation_history`	Read messages from a specific conversation.
`list_metrics`	List KPIs/metrics in a project.
`get_metric`	Retrieve full details and latest results for a metric.
`get_metric_results`	Access historical execution results for a metric.
`get_metric_insights`	View auto-detected anomalies and significant changes for a project's metrics.
`list_workflows`	List workflows defined in a project.
`get_workflow`	Retrieve workflow details and references.
`get_workflow_results`	Fetch results from a workflow's executions.
`list_artifacts`	List slides, dashboards, or reports in a project.
`get_artifact`	Fetch artifact content rendered as markdown.
`search_wiki`	Search the project knowledge base.
`list_recommendations`	Get AI-generated analysis suggestions for a project.

Setup Guide

Prerequisite: an Orion account with login credentials. The server is remote and requires no local installation.

Claude Desktop

Open Settings > Customize > Connectors, search for "Orion", click Add, then authenticate in the browser window that opens.

Claude Code

claude mcp add orion --transport http https://g.runorion.com/mcp

Then run /mcp inside Claude Code and complete the browser-based sign-in.

Cursor

Add an HTTP MCP server in Settings > MCP:

{
  "mcpServers": {
    "orion": {
      "type": "http",
      "url": "https://g.runorion.com/mcp"
    }
  }
}

Verify

Ask your assistant: "List my Orion projects." If accessible projects appear, the connection is working. Note that ask_orion analyses can take 1 to 2 minutes; the server polls automatically for results.

Full docs: https://docs.runorion.com/

Use Cases

Ask Orion to investigate why a KPI moved last week and have the narrative result delivered in Claude without opening the Orion app.
Pull the latest metric values and historical results into a planning doc by combining list_metrics, get_metric, and get_metric_results.
Surface anomalies across a project with get_metric_insights and have the agent draft a Slack-ready summary.
Trigger insight maker workflows (run_workflow) and collect results to generate weekly recurring reports.
Search the Orion knowledge base (search_wiki) for metric definitions or business context before answering analyst questions.

Example Prompts

"List my Orion projects, then show me the pinned metrics for the Growth project."
"Ask Orion why MRR dropped last week and wait for the full analysis."
"Pull the latest get_metric_insights for project X and summarize the top three anomalies."
"Run the weekly executive workflow in Orion and turn the results into a markdown briefing."
"Search the Orion wiki for our definition of 'qualified pipeline' and quote the relevant page."

Pros

Official, vendor-hosted remote MCP server, no local install or container to manage.
Broad tool coverage (20 tools) spanning analyses, metrics, workflows, artifacts, and knowledge base.
First-class support in Claude Desktop's connector marketplace, Claude Code, and Cursor with browser-based auth.
Designed for long-running analyses with automatic result polling.

Limitations

Requires a paid Orion account from Gravity; not useful without their underlying analytics platform.
Closed-source, provider-hosted: no public repo to inspect, fork, or self-host.
Multi-step analyses via ask_orion can take 1 to 2 minutes, which may feel slow in interactive use.

Alternatives

Looker MCP / Google's Looker connector for direct LookML-based querying.
Snowflake Cortex MCP server for in-warehouse analytics with an LLM.
dbt MCP server for exploring models, metrics, and lineage in a dbt project.