Databricks MCP Server – Governed Data and AI Tools
Databricks managed MCP servers connect AI agents to governed Databricks data and tools through Unity AI Gateway. Use them to query Genie Spaces, search AI Search indexes, run Databricks SQL, and invoke Unity Catalog functions while enforcing Unity Catalog permissions and centralized monitoring.
Overview
Databricks managed MCP servers are ready-to-use MCP endpoints inside a
Databricks workspace. They connect AI agents, coding assistants, and IDEs to
governed Databricks resources while Unity AI Gateway centralizes access control,
credential handling, and monitoring.
What the MCP server enables
Databricks provides several managed MCP server types. Depending on the endpoint
and the authenticated user's permissions, an AI agent can:
- Ask natural-language questions through Genie and poll for asynchronous
responses. - Query a single Genie Space for governed analysis of structured data.
- Search Databricks AI Search indexes that use Databricks managed embeddings.
- Run Databricks SQL for AI-assisted data work and pipeline authoring.
- Invoke Unity Catalog functions that execute predefined SQL logic.
- Access Unity Catalog tables, functions, and vector-style search resources
only when the user or service principal has permission. - Connect from clients such as Claude, Claude Code, Cursor, Replit, custom SDK
code, and other MCP-enabled tools.
When to use it
Use Databricks MCP when an AI agent needs live enterprise data context under
Databricks governance. Practical examples include answering business questions
with Genie, searching internal support documents through AI Search, running a
governed SQL analysis from an IDE, invoking approved Unity Catalog functions for
account lookups, or building an agent that combines structured tables,
unstructured indexes, and custom business logic.
Connection and authentication
Managed servers expose workspace-specific Streamable HTTP URLs. Common endpoint
patterns include /api/2.0/mcp/genie, /api/2.0/mcp/genie/{genie_space_id},
/api/2.0/mcp/ai-search/{catalog}/{schema}/{index_name},
/api/2.0/mcp/sql, and
/api/2.0/mcp/functions/{catalog}/{schema}/{function_name}.
OAuth is recommended because it supports scoped permissions and automatic token
refresh. Account admins create the OAuth application and configure client
redirect URLs. Personal access tokens are supported for managed and external
MCP servers and are suitable for individual development, testing, or short-term
access.
Key considerations
Databricks MCP is currently documented as Public Preview. Unity Catalog
permissions are always enforced, and users can only access servers and
underlying resources they are allowed to use. Databricks does not support
dynamic client registration for MCP OAuth flows, so external clients that
require dynamic registration are not supported through OAuth. Personal access
tokens are not supported for custom MCP servers hosted on Databricks Apps. If a
workspace uses IP access restrictions, the MCP client's outbound IP addresses
must be allowlisted. Long-running Genie and SQL operations require polling for
results, and compute pricing depends on the managed server type.
Supported Transports
streamable_http
URL: https://<workspace-hostname>/api/2.0/mcp/genie
streamable_http
URL: https://<workspace-hostname>/api/2.0/mcp/genie/{genie_space_id}
streamable_http
URL: https://<workspace-hostname>/api/2.0/mcp/ai-search/{catalog}/{schema}/{index_name}
streamable_http
URL: https://<workspace-hostname>/api/2.0/mcp/sql
streamable_http
URL: https://<workspace-hostname>/api/2.0/mcp/functions/{catalog}/{schema}/{function_name}
Frequently Asked Questions
- When should an AI agent use Databricks MCP servers?
- Use them when a workflow needs governed access to Databricks data or tools, such as asking Genie questions, searching AI Search indexes, running Databricks SQL, invoking Unity Catalog functions, or combining structured and unstructured enterprise data in an agent.
- What do Databricks MCP servers add to an AI agent's capabilities?
- They give the agent live, governed access to Databricks resources through Unity AI Gateway, enabling the agent to query permitted data and tools instead of relying only on static model knowledge or manually copied notebook and dashboard output.
- What can an AI agent access or manage through Databricks MCP?
- Depending on the managed server and permissions, the agent can access Genie conversations, individual Genie Spaces, AI Search indexes, Databricks SQL, and Unity Catalog functions. Unity Catalog permissions are enforced for the user or service principal.
- How is authentication configured for Databricks MCP servers?
- OAuth is recommended because it supports scoped permissions and token refresh. Account admins configure an OAuth application and redirect URLs. Personal access tokens can also be used for managed and external MCP servers, but custom MCP servers on Databricks Apps require OAuth.
- Which transport should be used for Databricks MCP servers?
- Use Streamable HTTP with the workspace-specific managed MCP endpoint URL. Databricks managed servers are remote workspace endpoints; no local stdio package is required for the managed Genie, AI Search, SQL, or Unity Catalog functions servers.