--- title: "Genie One MCP: How Agents Reach Unity Catalog Data Over MCP" url: "https://atlan.com/know/ai-agent/databricks/genie-one-mcp/" description: "How the Genie One MCP server lets agents query Unity Catalog data: tools, connection URL, OAuth, Unity Gateway grants and the other managed MCP servers." author: "Emily Winks" author_role: "Data Governance Expert" published: "2026-10-28" updated: "2026-10-28T00:00:00.000Z" --- > Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/ ## What is Genie One MCP? The Genie One MCP server is the Azure Databricks-provided Model Context Protocol (MCP) endpoint, system.ai.genie_one_mcp, that lets any MCP client or agent ask Genie a plain-language question and get an answer computed from your Unity Catalog data, with links to its sources. Calls run through Unity Gateway, and Unity Catalog permissions apply to every request. It became generally available on September 25, 2026. The server exposes five tools: - **genie_ask** starts a response to a question and returns conversation_id, response_id and status. - **genie_poll_response** fetches progress steps, the final answer and source links. - **genie_get_query_result** fetches the full SQL result, with column schema and rows. - **genie_cancel_response** cancels a turn that is still running. - **view_ask** asks a question and opens an interactive View, offered to MCP Apps clients. ## Key takeaways - The Genie One MCP answers questions across the whole workspace and is governed by Unity Gateway. - Unity Catalog permissions are enforced on every request, so results cover only what the caller can access. - The MCP server honors your Genie One configuration, so you tune answers in Genie One rather than in the agent. - Genie One and Genie Code read the same ontology, so a definition curated once serves both. --- The Genie One MCP server is the Azure Databricks-provided [Model Context Protocol (MCP)](https://atlan.com/know/mcp/why-mcp-matters-for-ai-agents/) endpoint, `system.ai.genie_one_mcp`, that lets any MCP client or agent ask Genie a plain-language question and get an answer computed from your Unity Catalog data, with links to its sources ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). Calls run through [Unity Gateway](https://atlan.com/know/ai-agent/databricks/unity-ai-gateway/), and Unity Catalog permissions apply to every request. It became generally available on September 25, 2026 ([September 2026 release notes](https://learn.microsoft.com/en-us/azure/databricks/release-notes/product/2026/september)). ### Plan Your Warehouse MCP Connection Takes your agent harness and your data warehouse, and returns the MCP server option, the authentication method, the permission scope, and the first read-only test to run. [Read the skill](/skills/warehouse-mcp-connection-plan.md). *Paste into a new chat* ``` Use the skill at https://atlan.com/skills/warehouse-mcp-connection-plan.md to plan how to connect our agent to Databricks over MCP. Ask me for whatever it needs. ``` *Run once in a terminal* ``` curl -fsSL --create-dirs \ -o ~/.agents/skills/warehouse-mcp-connection-plan/SKILL.md \ https://atlan.com/skills/warehouse-mcp-connection-plan.md ``` *For an agent* ``` curl -fsSL https://atlan.com/skills/warehouse-mcp-connection-plan.md ``` Human Agent ClaudeChatGPTGeminiCursorOther Copy --- ## What does the Genie One MCP server do when an agent calls it? Genie One is the Databricks agent that answers business questions across a workspace; [what Databricks Genie One is](https://atlan.com/know/ai-agent/databricks/genie-one/) covers the product. The MCP server [turns that agent into a tool](https://atlan.com/know/ai-agent/ai-agent-tool-use/). Your agent sends a question, and Genie searches your data, [writes SQL](https://atlan.com/know/ai-agent/data-for-ai/text-to-sql-for-enterprise/), runs it and returns an answer grounded in [Genie Ontology](https://atlan.com/know/ai-agent/databricks/genie-ontology/), with deep links to the Azure Databricks sources it cited ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). The server exposes five tools ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)): | Tool | What it does | |---|---| | `genie_ask` | Starts a response to a question and returns `conversation_id`, `response_id` and `status` | | `genie_poll_response` | Fetches progress steps, the final answer and source links | | `genie_get_query_result` | Fetches the full SQL result, with column schema and rows | | `genie_cancel_response` | Cancels a turn that is still running | | `view_ask` | Asks a question and opens an interactive View, offered to MCP Apps clients | Answers are asynchronous. `genie_ask` returns at once with `status: in_progress`, and the agent polls `genie_poll_response` until the status reads `completed`, `incomplete` or `failed`. A follow-up reuses the `conversation_id`, so the thread keeps its context ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). Two details shape agent design. `genie_ask` and `genie_poll_response` return truncated results to protect the model's context window, so an agent that needs every row calls `genie_get_query_result`, which still has a size cap ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). And clients that support the [MCP Apps extension](https://github.com/modelcontextprotocol/ext-apps/) show Genie's progress, charts and citations in a panel inside the client; other clients get text ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). Genie resolves business terms, [metric definitions](https://atlan.com/know/ai-agent/databricks/unity-catalog-metrics/) and table relationships through Genie Ontology first, which the docs say beats an agent [writing SQL against raw tables](https://atlan.com/know/ai-agent/semantic-layer/text-to-sql-vs-semantic-layer-for-self-serve-analytics/) ([Databricks-provided MCPs](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/built-in-mcp-services)). --- ## How do you connect an agent or MCP client to the Genie One MCP? Each Databricks-provided MCP sits in the `system.ai` schema as a Unity Catalog securable, so there is no server to host or register ([Databricks-provided MCPs](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/built-in-mcp-services)). The Genie One MCP URL carries its full Unity Catalog name and your workspace ID ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)): ```text https:///ai-gateway/mcp-services/system.ai.genie_one_mcp?o= ``` You find it under **Unity Gateway**, then **MCPs**. Setup depends on the client. - **[Coding agents](https://atlan.com/know/ai-agent/how-to-connect-ai-coding-agent-to-data-warehouse-mcp/).** The Unity Gateway CLI sets up Claude Code with `ug claude`, Codex with `ug codex` and Cursor's MCP access with `ug mcp add --agents cursor` ([Supported coding agents](https://learn.microsoft.com/en-us/azure/databricks/ai-gateway/coding-agent-supported-agents)). - **ChatGPT and Claude.** The Databricks Genie plugins connect both to the Genie One MCP through default OAuth apps. Users sign in with their Azure Databricks account, and no admin has to build an OAuth app ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). - **Python agents and apps.** Point the [MCP client](https://atlan.com/know/ai-agent/mcp-architecture-deep-dive/) at the URL above, as in [Call MCP tools from a Python agent](https://learn.microsoft.com/en-us/azure/databricks/agents/custom-agents/use-mcp-tools). For identity, the docs recommend [on-behalf-of user OAuth](https://atlan.com/know/ai-agent-access-control/) with the `ai-gateway` scope, so each answer is scoped to the person who asks. To pin queries to one SQL warehouse, pass `warehouse_id` in the `_meta` parameters of `genie_ask` or `view_ask`. For pricing, the server uses Chat in Genie One ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). An unattended agent can call as a service principal through OAuth machine-to-machine, and it then sees what that service principal can see ([MCP authentication](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/authentication-networking)). Databricks set October 31, 2026 as the sunset date for the earlier Beta endpoint, `https:///api/2.0/mcp/genie`; workloads on it must use `system.ai.genie_one_mcp` ([September 2026 release notes](https://learn.microsoft.com/en-us/azure/databricks/release-notes/product/2026/september)). --- ## Which managed MCP servers reach Unity Catalog data, and when do you use each? Azure Databricks offers several [MCP routes into Unity Catalog](https://atlan.com/know/mcp/mcp-server-for-databricks/). Status is as the docs state it today ([Databricks-provided MCPs](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/built-in-mcp-services)). | MCP route | Name or URL | Status in the docs | Governance | Use it for | |---|---|---|---|---| | Genie One MCP | `system.ai.genie_one_mcp` | Generally available | Unity Gateway grants and policies, plus data permissions | Business questions across the workspace | | Databricks SQL MCP | `system.ai.dbsql` | Needs the Unity Gateway beta, turned on by an account admin | Unity Gateway grants and policies, plus data permissions | A specific SQL query, such as validating syntax or writing a pipeline | | Genie Agent MCP | `/api/2.0/mcp/genie/{genie_space_id}` | Public Preview | Permissions on the Genie Agent and its data | Questions scoped to one curated Genie Agent | | AI Search MCP | `/api/2.0/mcp/ai-search/{catalog}/{schema}/{index_name}` | Public Preview | Permissions on the index | Retrieval from one search index | | Unity Catalog functions MCP | `/api/2.0/mcp/functions/{catalog}/{schema}/{function_name}` | Public Preview | Permissions on the function | A registered function called as a tool | A [Genie Agent](https://atlan.com/know/ai-agent/databricks/databricks-genie-context-requirements/) holds up to 50 Unity Catalog tables, views or [metric views](https://atlan.com/know/ai-agent/semantic-layer/how-to-create-databricks-metric-views/), so its MCP server is scoped to that curated set. It is read-only and passes no conversation history ([Genie Agent MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-agent)). The three `/api/2.0/mcp/` routes need a workspace admin to turn on **Managed MCP Servers** on the Previews page. Their calls bypass Unity Gateway, so they skip its tracing tables and take no MCP-level grants or policies. For new integrations, Databricks recommends the `system.ai` MCPs ([Databricks-provided MCPs](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/built-in-mcp-services)). A third-party server is a separate path: you register it in Unity Gateway through a Unity Catalog HTTP connection, and it must support [Streamable HTTP](https://atlan.com/know/mcp-server-implementation-guide/) ([External MCP servers](https://learn.microsoft.com/en-us/azure/databricks/ai-gateway/register-mcp-service)). --- ## How does Unity Catalog control what an agent sees through MCP? Two gates apply to a Genie One MCP call. The first decides who may call the MCP at all. A caller needs `EXECUTE` on the MCP, `USE CATALOG` and `USE SCHEMA` on its parents, and assignment to the workspace. The `system.ai` MCPs can inherit a default `EXECUTE` grant on `system.ai`, so review those broad grants if you want to restrict them. To grant access, a metastore admin first grants themselves `MANAGE` on the `system.ai` schema ([Govern an MCP](https://learn.microsoft.com/en-us/azure/databricks/ai-gateway/govern-mcp-service)). The second gate is the data. [Unity Catalog permissions](https://atlan.com/know/ai-agent/context-layer/context-layer-role-based-access-control/) are enforced on every request, so results cover only what the caller can access ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). The context is gated too: Genie uses only [ontology snippets](https://atlan.com/know/ai-agent/ontology/how-agents-use-ontology/) taken from assets the caller has permission to see ([Genie Ontology](https://learn.microsoft.com/en-us/azure/databricks/genie/genie-ontology)). Unity Gateway adds controls on top ([Govern an MCP](https://learn.microsoft.com/en-us/azure/databricks/ai-gateway/govern-mcp-service)): - **Policies, in Beta.** A policy checks a tool call before it runs and can check its result. It can [allow, block or require approval](https://atlan.com/know/ai-agent-risks-guardrails/). The built-in [Sensitive Data Detection policy](https://atlan.com/know/ai-agent/data-for-ai/how-to-handle-pii-in-ai-pipelines/) blocks tagged categories such as `class.us_ssn` in inputs and outputs. [External guardrail policies](https://atlan.com/know/ai-agent/ai-guardrails-tools-compared/) fail closed: if the vendor endpoint is unreachable, the call is denied. - **[Rate limits](https://atlan.com/know/what-is-ai-gateway-llm-gateway/).** Caps on `tools/call` requests in queries per minute; calls over the cap return HTTP 429. - **[Monitoring](https://atlan.com/know/ai-agent-observability/).** Call volume in `system.ai_gateway.usage` with `service_type = 'MCP_SERVICE'`, and invocations logged as `mcpCall` in `system.access.audit`. Routing, authentication, rate limits and observability in front of MCP servers are the jobs of [an MCP gateway](https://atlan.com/know/ai-agent/what-is-an-mcp-gateway/), and Unity Gateway does them for Databricks workspaces. For when a gateway pays off, see [MCP gateway vs. single MCP server](https://atlan.com/know/ai-agent/mcp-gateway-vs-single-mcp-server/). --- ## What context does Genie One use to answer, and where does it come from? Genie One answers from Genie Ontology, which is in Public Preview. The ontology joins two kinds of context ([Genie Ontology](https://learn.microsoft.com/en-us/azure/databricks/genie/genie-ontology)): - **[Unity Catalog semantics](https://atlan.com/know/ai-agent/semantic-layer/unity-catalog-semantic-layer/).** Context you model, govern and certify, through metric views, domains and Pages. - **Inferred context.** Snippets Genie extracts from your metric views, dashboards, SQL queries and Genie Agents. Queries that external tools such as dbt run on Azure Databricks count, because they appear in query history. Each snippet carries an authority score built from its source, how often it is used and [how fresh it is](https://atlan.com/know/ai-agent/context-freshness/). At question time, Genie ranks the relevant snippets, resolves conflicts and answers from the sources the caller may see. Genie One and [Genie Code](https://atlan.com/know/ai-agent/databricks/databricks-data-ai-summit-2026-announcements/) read the same ontology, so a definition curated once serves both. The MCP server honors your Genie One configuration, so you tune answers in Genie One rather than in the agent ([Genie One MCP server](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp)). Workspace admins can add instructions in `/Workspace/.genie_workspace_instructions.md`, under 20,000 characters, and chat can draw on connected sources such as Google Drive and SharePoint ([Chat in Genie One](https://learn.microsoft.com/en-us/azure/databricks/genie-one/chat)). The same doc warns that many Genie Agents in one workspace can lower routing accuracy. Both kinds of context come from Unity Catalog and from assets and queries in the workspace, so the answer an agent gets over MCP is as good as the metric views, Genie Agents and queries behind it ([Genie Ontology](https://learn.microsoft.com/en-us/azure/databricks/genie/genie-ontology)). --- ## How does Atlan work with Databricks Genie and Unity Catalog over MCP? Atlan plugs into the same MCP layer from the other side. [Atlan's MCP server](https://atlan.com/know/what-is-atlan-mcp/) installs from the Databricks Marketplace: an admin enters an Atlan OAuth client during install, each user signs in to Atlan through the new connection in Catalog, and the server can then be added as a tool in AI Playground ([Set up Atlan MCP for Databricks](https://docs.atlan.com/product/capabilities/atlan-ai/how-tos/remote-mcp-overview?client=databricks)). Atlan's release notes describe this install as the way to [use Atlan MCP with Databricks Genie](https://shipped.atlan.com/atlan-mcp-faster-safer-everywhere-22TNJK), with every interaction scoped to the connected user's Atlan permissions. [Atlan MCP exposes 29 tools across 8 categories](https://atlan.com/mcp/). Atlan also catalogs Genie Agents. Its [Databricks connector](https://atlan.com/know/context-layer-for-databricks/) crawls each Genie space as a Genie Agent asset and [draws lineage](https://atlan.com/know/ai-agent/data-for-ai/data-lineage-for-ai/) from the Unity Catalog tables, views and columns it reads, using `system.access.table_lineage` and `system.access.column_lineage`. This support is in private preview ([How Atlan builds lineage for Databricks Genie Agents](https://docs.atlan.com/apps/connectors/data-warehouses/databricks/faq/how-databricks-genie-lineage-works)). It shows which tables sit behind each Genie Agent. Workday puts its [shared business language](https://atlan.com/know/ai-agent/how-to-manage-kpi-definitions-and-a-business-glossary-in-a-data-catalog/) behind Atlan's MCP server. Joe DosSantos, VP Enterprise Data & Analytics at Workday: "Atlan captures Workday's shared language to be leveraged by AI via its MCP server. As part of Atlan's AI labs, we're co-building the semantic layer that AI needs." ([Atlan MCP](https://atlan.com/mcp/)) For how Atlan context and Genie Ontology fit together, read [Genie Ontology and the Atlan context layer](https://atlan.com/know/ai-agent/databricks/genie-ontology-and-atlan-context-layer/) and [how to add enterprise context to a Databricks Genie agent](https://atlan.com/know/ai-agent/databricks/add-context-to-databricks-genie-space/). --- ## FAQs about Genie One MCP ### Is the Genie One MCP server generally available? Yes. The Azure Databricks release notes list the Genie One MCP server as generally available on September 25, 2026, as the Databricks-provided MCP `system.ai.genie_one_mcp` in Unity Gateway. Releases are staged, so a given account can receive a feature a week or more after the first release date. ### What is the difference between the Genie One MCP and the Genie Agent MCP? The Genie One MCP answers questions across the whole workspace and is governed by Unity Gateway. The Genie Agent MCP queries one curated Genie Agent of up to 50 tables, views or metric views, bypasses Unity Gateway and is in Public Preview. Use the Genie Agent MCP when the agent should stay inside one domain. ### Can an agent see data it has no permission for through Genie One MCP? No. Unity Catalog permissions are enforced on every request, and ontology snippets are gated by the same permissions. With on-behalf-of user OAuth, the agent sees what the signed-in user can see. With a service principal, it sees what that principal can see, so scope its grants tightly. ### Which clients can connect to the Genie One MCP server? Any client that speaks MCP over the Unity Gateway URL. The docs give setup paths for Claude Code, Codex, Cursor, Gemini CLI and GitHub Copilot CLI through the `ug` CLI, Databricks Genie plugins for ChatGPT and Claude, and Python agents. Clients that support MCP Apps also get the interactive View. --- ## Sources 1. [Genie One MCP server, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-mcp) 2. [September 2026 release notes, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/release-notes/product/2026/september) 3. [Databricks-provided MCPs, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/built-in-mcp-services) 4. [Supported coding agents, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/ai-gateway/coding-agent-supported-agents) 5. [Call MCP tools from a Python agent, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/agents/custom-agents/use-mcp-tools) 6. [MCP authentication, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/authentication-networking) 7. [Genie Agent MCP server, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/agents/mcp-tools/genie-agent) 8. [External MCP servers, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/ai-gateway/register-mcp-service) 9. [Govern an MCP, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/ai-gateway/govern-mcp-service) 10. [Genie Ontology, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/genie/genie-ontology) 11. [Chat in Genie One, Azure Databricks documentation on Microsoft Learn](https://learn.microsoft.com/en-us/azure/databricks/genie-one/chat) 12. [Set up Atlan MCP, Atlan Documentation](https://docs.atlan.com/product/capabilities/atlan-ai/how-tos/remote-mcp-overview?client=databricks) 13. [How Atlan builds lineage for Databricks Genie Agents, Atlan documentation](https://docs.atlan.com/apps/connectors/data-warehouses/databricks/faq/how-databricks-genie-lineage-works)