Omni Analytics MCP Server

Omni's MCP server connects external AI clients (Claude Desktop, ChatGPT, Cursor, VS Code, Codex) to your Omni instance so users can ask natural language questions about their data while results stay governed by Omni's semantic model. Because queries flow through the semantic layer, AI responses inherit the same access controls, business definitions, and security policies that already exist in Omni, which helps prevent hallucinations and data leakage compared to raw text-to-SQL.

The server exposes a small set of tools that guide an LLM through a query pipeline: choose a model, optionally narrow to a topic, then execute a query and return results. It also exposes a documentation search tool for grounding answers in official Omni docs. Sessions support multi-step iterative analysis with memory retention, and scope can be locked to a specific model or topic so the AI cannot wander outside permitted data.

The server is hosted on your Omni instance at https://<YOUR_OMNI_INSTANCE>/mcp/https and is reached either through a direct HTTP/SSE connection or via the @omni-co/mcp npm bridge for stdio-based clients. Authentication supports OAuth 2.1 with PKCE (recommended) or a manually issued Omni API key.

Tools

Prerequisites

• An Omni account and instance (e.g. acme.omniapp.co)
• The MCP server feature enabled under Settings > AI > General in Omni
• For OAuth: Personal Access Tokens enabled for your workspace
• For API key auth: an Omni API key created at Settings > API Keys

Authentication options

• OAuth 2.1 (recommended): Browser-based flow with PKCE. Connection URL: https://callbacks.omniapp.co/callback/mcp. Any user (including Viewers) can initiate it; permissions still apply.
• API key: Manual setup. Connection URL: https://<YOUR_OMNI_INSTANCE>/mcp/https.

Claude Desktop config (API key method)

1. Create an Omni API key.
2. Open Claude Desktop, go to Settings > Developer, click Edit Config.
3. Add the following to claude_desktop_config.json:

{
  "mcpServers": {
    "Omni MCP": {
      "command": "npx",
      "args": ["@omni-co/mcp"],
      "env": {
        "DEBUG": "true",
        "MCP_SERVER_URL": "https://<YOUR_OMNI_INSTANCE>/mcp/https",
        "MCP_API_KEY": "<OMNI_API_KEY>",
        "MCP_MODEL_ID": "<OPTIONAL_OMNI_MODEL_ID>",
        "MCP_TOPIC_NAME": "<OPTIONAL_TOPIC_NAME>",
        "MCP_USER_REQUIRED": "<true/false>",
        "MCP_USER_ID": "<USER_ID>",
        "MCP_QUERY_ALL_VIEWS": "<true/false>"
      }
    }
  }
}

4. Replace <YOUR_OMNI_INSTANCE> and <OMNI_API_KEY> with your real values.
5. Save and restart Claude Desktop. Check Settings > Developer for a "running" badge.

Supported clients

Omni publishes setup guides for ChatGPT, Claude Code, Claude Desktop, Codex, Cursor, and Visual Studio Code at https://docs.omni.co/ai/mcp.

• Let business users ask natural language questions about KPIs in Claude or ChatGPT and get answers backed by Omni's semantic model rather than raw SQL guesses.
• Chain a multi-step analysis (pickModel, pickTopic, getData) inside Cursor or Claude Code to investigate revenue, funnel, or product metrics without leaving the IDE.
• Scope an AI agent to a single Omni model or topic so it can only query a specific domain (e.g. marketing analytics) while inheriting user-level row and field permissions.
• Pull governed data into a working document by combining Omni MCP with Google Docs or Notion MCP, e.g. a product marketer summarizing recent won deals for positioning updates.
• Use searchOmniDocs to answer modeling and admin questions ("how do I define a measure?") with cited Omni documentation.

• "List the available Omni models and show me the topics in the 'Sales Analytics' model."
• "What was weekly active users by plan tier for the last 8 weeks? Use the product_analytics model."
• "Find the top 10 accounts by closed-won ARR this quarter and break down by industry."
• "Compare conversion rates across acquisition channels for the last 30 days, then drill into the lowest performer."
• "Search the Omni docs for how to define a row-level access filter and summarize the steps."

• Official, vendor-maintained server with first-party support for major MCP clients (Claude Desktop, ChatGPT, Cursor, VS Code, Codex, Claude Code).
• Queries inherit Omni's semantic model and user-level permissions, which reduces hallucinations and prevents unauthorized data access.
• Supports OAuth 2.1 with PKCE as well as API key auth, enabling both end-user and service-account use cases.
• Scoping options (model ID, topic, query-all-views header) let admins lock an agent to a specific data domain.

• Requires an Omni subscription and an Omni instance; not usable without an existing Omni deployment.
• MCP OAuth PATs are not currently visible in the Omni UI and must be retrieved through the API tokens endpoint.
• Functionality is limited to read/query operations against modeled data; it is not a general admin or write API for Omni.

• Hex MCP for notebook-based analytics workflows over warehouse data.
• Cube MCP for semantic-layer querying when using Cube as the metrics layer.
• MCP Toolbox for Databases (Google) for direct, ungoverned SQL access to warehouses like BigQuery and Postgres.