> ## Documentation Index
> Fetch the complete documentation index at: https://docs.omni.co/llms.txt
> Use this file to discover all available pages before exploring further.

# AI MCP Server

> Unlock new ways to analyze & explore your data by connecting external AI tools to your Omni model with Omni's MCP Server.

The Omni MCP Server enables natural language querying across your datasets, similar to how you’d interact with Omni, while taking advantage of the enhanced context, reasoning, or memory capabilities of external AI tools.

For example, you could include the results of a web search as additional context to the MCP before running an Omni query.

<img src="https://mintcdn.com/omni-e7402367/t-IfvrVPwJe-flyV/ai/images/mcp-in-cursor.png?fit=max&auto=format&n=t-IfvrVPwJe-flyV&q=85&s=a822d8f1303b46d317a507a726a998a9" alt="MCP Server in Cursor" width="1730" height="1081" data-path="ai/images/mcp-in-cursor.png" />

## Features

* **Natural language querying** – Ask questions using natural language and let Omni translate it into a query
* **Iterative, multi-step analysis** – Chain queries together, retain results in memory, and perform deeper analysis across steps
* **Flexible integration** - Connect to a variety of compatible AI clients to expand beyond Omni's native chat experience
* **Dynamic model selection** - Switch between Omni models during a session without reconfiguring the MCP Server
* **Scoped, secure data access** – Interactions are scoped only to the model (and topic, if provided) provided to the MCP, preventing unauthorized data access when querying
* **User level permissioning** - MCP queries can be scoped to the permissions of a given user to pass through security from Omni's semantic layer

## Tools

The MCP Server includes tools for querying Omni data and searching the official Omni docs for product and *"How do I..."*-type questions.

### Querying

When you send a query to the MCP Server, it uses the following tools to process your request:

<Steps>
  <Step title="pickModel" noAnchor>
    The `pickModel` tool returns a list of available models and their IDs. The MCP client uses this to select which model to query. You can also provide a model ID in the configuration to skip this step.
  </Step>

  <Step title="pickTopic" noAnchor>
    Once a model is selected, this tool returns the available topics within that model. The MCP client selects the most relevant topic for your query.

    If the AI is allowed to query any view in the model (enabled with the `X-MCP-Query-All-Views` header during setup), this step becomes optional: the AI can query any view in the model directly, not just topics.
  </Step>

  <Step title="getData" noAnchor>
    The `getData` tool executes the query against the selected model and topic (or view, if the AI can query outside of topics), returning the results.
  </Step>
</Steps>

This dynamic selection flow allows you to switch between models during a session without needing to reconfigure the MCP Server.

### Documentation search

In addition to data querying tools, the MCP Server includes a **searchOmniDocs** tool that provides AI-powered search of Omni's documentation.

When you ask questions about how to use Omni, the MCP client can automatically use this tool to:

* Search Omni's documentation for relevant information
* Synthesize answers from multiple documentation sources
* Provide citations to specific documentation pages

For example, you could ask *"How do I create a calculated field in Omni?"* and the **searchOmniDocs** tool will search the documentation, analyze the relevant pages, and return a comprehensive answer with links to the source documentation.

The tool uses the same AI-powered search pipeline as the Omni Agent, ensuring accurate and contextually relevant responses.

## Authentication

The Omni MCP Server supports two authentication methods:

* **OAuth 2.1** - **Recommended**. Streamlined setup through your browser. Omni creates API keys on your behalf automatically. Requires [**Personal Access Tokens (PATs)**](/api/authentication#personal-access-tokens-pat) to be enabled for your organization.
* **API key** - Manual key creation for centralized management or automated workflows.

See the [MCP authentication guide](/ai/mcp/authentication) for more information.

## Data access and security

* **Data access** - Interactions are scoped only to the model (and topic, if provided) provided to the MCP, preventing unauthorized data access when querying
* **User permissions** - All MCP queries use the permissions of the authenticated user, ensuring proper data access controls
* **Data handling & security** - Data returned from queries run by the MCP is handled the same way it is for [Omni's other AI features](/ai/security). **Note**: By using the MCP Server, your data will be in the tool where you use the MCP Server. For example, if you use Cursor with the MCP Server, your data would be in Cursor.

## Getting started

Ready to get started? Check out the setup guides for the clients we've tested for compatibility:

<Columns cols={3}>
  <Card title="ChatGPT" href="/ai/mcp/chatgpt" icon="message-bot">
    OpenAI's AI assistant with developer mode
  </Card>

  <Card title="Claude Code" href="/ai/mcp/claude-code" icon="code">
    Command-line based version of Anthropic Claude
  </Card>

  <Card title="Claude Desktop" href="/ai/mcp/claude-desktop" icon="desktop">
    App version of Anthropic Claude
  </Card>

  <Card title="Codex" href="/ai/mcp/codex" icon="terminal">
    OpenAI's agentic coding tool
  </Card>

  <Card title="Cursor" href="/ai/mcp/cursor" icon="arrow-pointer">
    AI-powered IDE designed for fast, agentic programming
  </Card>

  <Card title="Visual Studio Code" href="/ai/mcp/vscode" icon="microsoft">
    Popular open-source IDE with powerful AI features
  </Card>
</Columns>

## Improving results

To get the best results, we recommend:

1. Optimizing your Omni models for AI. Refer to [this guide](/modeling/develop/ai-optimization) to learn how to curate your models and include context.
2. Rating the results that the AI returns