Databricks
Databricks is the lakehouse company, helping data teams solve the world's toughest problems with unified analytics platform for big data and AI.
Verdict
Common use cases
- Cancel runaway jobs from incident chat
- Grant metastore access during onboarding
- Check pipeline status at daily standup
- Provision external storage connections on demand
- Audit group memberships for compliance reviews
Integration
- Vendor
- Databricks
- Category
- developer-tools
- Auth
- OAUTH2
- Tools
- 50
- Composio slug
databricks
Tools
- Add Member to Security Group
Tool to add a user or group as a member to a Databricks security group. Use when you need to grant group membership for access control.
- Assign Metastore to Workspace
Tool to assign a Unity Catalog metastore to a workspace. Use when you need to link a workspace to a Unity Catalog metastore, enabling shared data access with consistent governance policies. Requires account admin privileges. If an assignmen
- Cancel Databricks Job Run
Tool to cancel a Databricks job run asynchronously. Use when you need to terminate a running job. The run will be terminated shortly after the request completes. If the run is already in a terminal state (TERMINATED, SKIPPED, or INTERNAL_ER
- Create Catalog Connection
Tool to create a new Unity Catalog connection to external data sources. Use when you need to establish connections to databases and services such as MySQL, PostgreSQL, Snowflake, etc. Requires metastore admin privileges or CREATE CONNECTION
- Create Catalog Credential
Tool to create a new credential for Unity Catalog access to cloud services. Use when you need to establish authentication for STORAGE (cloud storage) or SERVICE (external services like AWS Glue) purposes. Requires metastore admin or CREATE_
- Create Databricks App
Tool to create a new Databricks app with specified configuration. Use when you need to create apps hosted on Databricks serverless platform to deploy secure data and AI applications. The app name must be unique within the workspace, contain
- Create External Location
Tool to create a new Unity Catalog external location combining a cloud storage path with a storage credential. Use when you need to establish access to cloud storage in Azure Data Lake Storage, AWS S3, or Cloudflare R2. Requires metastore a
- Create Metastore
Tool to create a new Unity Catalog metastore. Use when you need to establish a top-level container for data in Unity Catalog, registering metadata about securable objects (tables, volumes, external locations, shares) and access permissions.
- Delete Catalogdestructive
Tool to delete a catalog from Unity Catalog metastore. Use when you need to permanently remove a catalog and optionally its contents. By default, the catalog must be empty (except for information_schema). Use force=true to delete non-empty
- Delete Catalog Connectiondestructive
Tool to delete a Unity Catalog connection to external data sources. Use when you need to remove connections to databases and services. Deleting a connection removes the abstraction used to connect from Databricks Compute to external data so
- Delete Catalog Credentialdestructive
Tool to delete a Unity Catalog credential for cloud storage or service access. Use when you need to remove credentials that authenticate access to cloud resources. By default, deletion will fail if the credential has dependent resources. Us
- Delete Custom LLM Agentdestructive
Tool to delete a Custom LLM agent created through Agent Bricks. Use when you need to remove a custom LLM and all associated data. This operation is irreversible and deletes all data including temporary transformations, model checkpoints, an
- Delete Databricks Appdestructive
Tool to delete a Databricks app from the workspace. Use when you need to remove an app and its associated service principal. When an app is deleted, Databricks automatically deletes the provisioned service principal.
- Delete External Locationdestructive
Tool to delete an external location from Unity Catalog metastore. Use when you need to remove an external location that combines a cloud storage path with a storage credential. The caller must be the owner of the external location. Use forc
- Delete Metastoredestructive
Tool to delete a Unity Catalog metastore. Use when you need to permanently remove a metastore and its managed data. Before deletion, you must delete or unlink any workspaces using the metastore. All objects managed by the metastore will bec
- Delete Online Tabledestructive
Tool to delete an online table by name. Use when you need to permanently remove an online table and stop data synchronization. This operation deletes all data in the online table permanently and releases all resources. Note: online tables a
- Deploy Databricks App
Tool to create a deployment for a Databricks app. Use when you need to deploy an app with source code from a workspace path. The deployment process provisions compute resources and uploads the source code. Deployments can be in states: IN_P
- Generate Temporary Service Credential
Tool to generate temporary credentials from a service credential with admin access. Use when you need short-lived, scoped credentials for accessing cloud resources. The caller must be a metastore admin or have the ACCESS privilege on the se
- Get App Deployment Update
Tool to retrieve information about a specific app deployment update. Use when you need to track the status and details of app deployment updates, including whether the update succeeded, failed, is in progress, or was not updated.
- Get Catalog Artifact Allowlist
Tool to retrieve artifact allowlist configuration for a specified artifact type in Unity Catalog. Use when you need to check which artifacts are permitted for use in your Databricks environment. Requires metastore admin privileges or MANAGE
- Get Catalog Connection
Tool to retrieve detailed information about a specific Unity Catalog connection. Use when you need to get connection metadata, configuration, and properties for external data source connections.
- Get Catalog Credential
Tool to retrieve detailed information about a specific Unity Catalog credential. Use when you need to get credential metadata, configuration, and cloud provider details for storage or service credentials.
- Get Catalog Details
Tool to retrieve details of a specific catalog in Unity Catalog. Use when you need to get information about a catalog including its metadata, owner, properties, and configuration. Requires metastore admin privileges, catalog ownership, or U
- Get Catalog Grants
Tool to get permissions (grants) for a securable in Unity Catalog without inherited permissions. Use when you need to see direct privilege assignments on a catalog or other securable object. Returns only privileges directly assigned to prin
- Get Current Metastore Assignment
Tool to retrieve the current metastore assignment for the workspace being accessed. Use when you need to determine which metastore is assigned to the current workspace context.
- Get Databricks App Details
Tool to retrieve details about a specific Databricks app by name. Use when you need to get comprehensive information about an app including configuration, deployment status, compute resources, and metadata.
- Get Databricks App Permission Levels
Tool to retrieve available permission levels for a Databricks app. Use when you need to understand what permission levels can be assigned to users or groups for a specific app. Returns permission levels like CAN_USE and CAN_MANAGE with thei
- Get Databricks App Permissions
Tool to retrieve permissions for a Databricks app. Use when you need to check who has access to an app and their permission levels. Returns the access control list including inherited permissions from parent or root objects.
- Get Effective Catalog Permissions
Tool to get effective permissions for a securable in Unity Catalog, including inherited permissions from parent securables. Use when you need to understand what privileges are granted to principals through direct assignments or inheritance.
- Get Entity Tag Assignment
Tool to retrieve a specific tag assignment for a Unity Catalog entity by tag key. Use when you need to get details about a tag assigned to catalogs, schemas, tables, columns, or volumes. Requires USE CATALOG and USE SCHEMA permissions on pa
- Get External Location Details
Tool to retrieve details of a specific Unity Catalog external location. Use when you need to get information about an external location including its URL, storage credential, and configuration. Requires metastore admin privileges, external
- Get Metastore Details
Tool to retrieve detailed information about a specific Unity Catalog metastore by its ID. Use when you need to get metastore configuration, ownership, storage settings, and Delta Sharing details. Requires metastore admin permissions.
- Get Metastore Summary
Tool to retrieve summary information about the metastore associated with the current workspace. Use when you need metastore configuration overview including cloud vendor, region, storage, and Delta Sharing details.
- Get Model Version
Tool to retrieve detailed information about a specific version of a registered model in Unity Catalog. Use when you need to get metadata, status, source location, and configuration of a model version. Requires metastore admin privileges, mo
- Set Databricks App Permissions
Tool to set permissions for a Databricks app, replacing all existing permissions. Use when you need to configure access control for an app. This operation replaces ALL existing permissions; for incremental updates, use the update permission
- Start Databricks App
Tool to start the last active deployment of a Databricks app. Use when you need to start a stopped app, which transitions it to the ACTIVE state. The start operation is asynchronous and the app will transition to the ACTIVE state after the
- Stop Databricks App
Tool to stop the active deployment of a Databricks app. Use when you need to stop a running app, which transitions it to the STOPPED state. The stop operation is asynchronous and the app will transition to the STOPPED state after the operat
- Unassign Metastore from Workspace
Tool to unassign a Unity Catalog metastore from a workspace. Use when you need to remove the association between a workspace and its assigned metastore, leaving the workspace with no metastore. The metastore itself is not deleted, only the
- Update Catalog Connection
Tool to update an existing Unity Catalog connection configuration. Use when you need to modify connection properties, credentials, ownership, or metadata for external data sources.
- Update Catalog Credential
Tool to update an existing Unity Catalog credential with new properties. Use when you need to modify credential configuration, ownership, or cloud provider settings. The caller must be the owner of the credential, a metastore admin, or have
- Update Catalog Function
Tool to update function owner in Unity Catalog. Use when you need to change the ownership of a catalog function. Only the owner of the function can be updated via this endpoint. The caller must be a metastore admin, the owner of the functio
- Update Catalog Grants
Tool to update permissions for Unity Catalog securables by adding or removing privileges for principals. Use when you need to grant or revoke permissions on catalogs, schemas, tables, or other Unity Catalog objects. Only metastore admins, o
- Update Databricks App
Tool to update an existing Databricks app configuration. Use when you need to modify app settings such as description, resources, compute size, budget policy, or API scopes. This is a partial update operation - only fields provided in the r
- Update Databricks App Permissions
Tool to incrementally update permissions for a Databricks app. Use when you need to modify specific permissions without replacing the entire permission set. This PATCH operation updates only the specified permissions, preserving existing pe
- Update External Location
Tool to update an existing Unity Catalog external location properties. Use when you need to modify the cloud storage path, credentials, ownership, or configuration of an external location. The caller must be the owner of the external locati
- Update External Metadata
Tool to update an external metadata object in Unity Catalog. Use when you need to modify metadata about external systems registered within Unity Catalog. The user must have metastore admin status, own the object, or possess the MODIFY privi
- Update Metastore
Tool to update configuration settings for an existing Unity Catalog metastore. Use when you need to modify metastore properties like name, owner, Delta Sharing settings, or storage credentials. Requires metastore admin permissions.
- Update Metastore Assignment
Tool to update a metastore assignment for a workspace. Use when you need to update the metastore_id or default_catalog_name for a workspace that already has a metastore assigned. Account admin privileges are required to update metastore_id,
- Update Model Version
Tool to update a Unity Catalog model version. Use when you need to modify the comment of a specific model version. Currently only the comment field can be updated. The caller must be a metastore admin or owner of the parent registered model
- Validate Catalog Credential
Tool to validate a Unity Catalog credential for external access. Use when you need to verify that a credential can successfully perform its intended operations. For SERVICE credentials, validates cloud service access. For STORAGE credential
Setup
Setup guide
- 11. In Switchy, open your Space settings and click 'Add Integration'. 2. Search for Databricks and select it from the developer tools category. 3. Click 'Connect' to start the OAuth flow — you'll be redirected to Databricks. 4. Sign in with an account that has workspace admin permissions (required for most tools). 5. Review the requested scopes: workspace management, job execution, Unity Catalog administration, and group membership. 6. Authorize the connection and you'll return to Switchy. 7. Test it by typing '@Databricks list all running jobs' in your Space — you should see active runs or a confirmation that none are running. 8. To invoke any tool, @mention Databricks and describe what you need in plain language.
What teammates see: by default, memories from Databricks are scoped to the Space (PROJECT visibility) - you can mark any memory PRIVATE or share it ORG-wide.
Works well with
Top models
Compatibility data appears once enough Spaces have used this MCP together with a given model.
How Switchy teams use it
Starter prompts
Cancel Stuck Job
@Databricks cancel the job run with ID 47382. It's been stuck for two hours and blocking our ETL pipeline.Open in a Space →
Add User to Group
@Databricks add sarah@company.com to the data-engineers security group so she can access production tables.Open in a Space →
Create External Location
@Databricks create an external location named 'marketing-data' pointing to s3://company-marketing-bucket using the existing aws-prod credential.Open in a Space →
List Running Jobs
@Databricks show me all currently running jobs in the workspace, including their start times and cluster IDs.Open in a Space →
Assign Metastore
@Databricks assign the 'prod-metastore' Unity Catalog metastore to workspace ID ws-9f3a2b1c so the team can query shared tables.Open in a Space →
Example outputs
Illustrative - representative of the model's voice and quality, not literal recordings.
@databricks list all active job runs in the analytics workspace and show which ones have been running longer than 2 hours
Based on the current workspace state, there are 4 active job runs. Two have exceeded the 2-hour threshold: 'daily-customer-segmentation' (started 3h 12m ago, currently processing 847M records) and 'real-time-fraud-detection' (started 2h 41m ago, in model training phase). The other two runs are within normal execution time. This example illustrates how the MCP would surface job metadata — actual run IDs, timestamps, and cluster states would appear in a real query.
This read-heavy example demonstrates the MCP's monitoring capabilities across Databricks jobs. Requires workspace-level read permissions. Useful for spotting long-running jobs that may need intervention, though the MCP won't auto-cancel — you'd need a follow-up prompt to trigger the 'Cancel Databricks Job Run' tool if needed.
@databricks create a new external location called 'marketing-data-lake' pointing to s3://acme-marketing-prod/datasets using the existing prod-s3-credential
External location 'marketing-data-lake' created successfully in Unity Catalog. The location is now registered at s3://acme-marketing-prod/datasets with storage credential 'prod-s3-credential'. Teams with appropriate grants can now create tables and volumes referencing this path. This example shows the confirmation you'd receive — the actual response includes the location's unique ID and initial permission state.
This write example showcases Unity Catalog governance setup. Requires metastore admin privileges and assumes the storage credential already exists. Side-effect: immediately makes the S3 path accessible to users with grants on this location. Always verify the credential has correct IAM policies before creating the location to avoid permission errors downstream.
@databricks review all security groups in the workspace, identify which ones have more than 20 members, and draft a governance report recommending consolidation
Analyzed 18 security groups across the workspace. Five groups exceed 20 members: 'data-engineers' (34 members), 'ml-researchers' (28), 'analytics-viewers' (45), 'prod-deployers' (23), and 'legacy-admin-access' (31). The 'legacy-admin-access' group warrants immediate attention — it grants cluster creation rights to 31 users, many of whom haven't logged in for 90+ days. Recommend consolidating 'analytics-viewers' and 'data-engineers' into role-based groups aligned to current project teams. This illustrative synthesis combines group membership data with AI reasoning about access patterns.
This synthesis example pairs the MCP's group enumeration tools with the AI's ability to spot governance anti-patterns. Requires account-level read access to enumerate all groups and members. The MCP provides raw membership data; the AI infers risk based on naming conventions and member counts. Use this to audit sprawling permission structures, but always validate recommendations with your security team before making changes.
Use-case deep-dives
When Databricks MCP speeds up analyst access provisioning
A 12-person analytics team at a fintech startup needs to onboard three new analysts who require read access to production tables and write access to a staging schema. The Databricks MCP's security group and metastore assignment tools let you script the entire provisioning flow in a single Switchy prompt: add users to the correct groups, assign the workspace metastore, and verify external location permissions. This works well up to about 30 users per workspace—beyond that threshold, you're better off writing a dedicated provisioning script because the MCP's per-call latency adds up. If your onboarding cadence is monthly or slower and your team is under 50 people, the MCP saves you from context-switching into the Databricks console every time someone joins.
Why this MCP falls short for real-time job monitoring
A 6-person data engineering team runs 40 nightly ETL jobs and needs to triage failures during their 9am standup. The Databricks MCP includes a job-run cancellation tool, but it lacks read-only monitoring tools like listing recent runs or fetching error logs. You can cancel a stuck job from Switchy, but you can't see which jobs failed or why without opening the Databricks UI. This makes the MCP a poor fit for standup triage workflows where the team needs to scan job statuses and drill into logs quickly. If your workflow is primarily reactive monitoring rather than proactive job management, the MCP won't save you time—stick with the native Databricks console or a dedicated observability tool.
When this MCP wins for multi-cloud Unity Catalog config
A 5-person platform team is migrating from a single AWS deployment to a multi-cloud architecture spanning AWS S3 and Azure Data Lake. They need to create external locations, storage credentials, and catalog connections for each cloud provider. The Databricks MCP's credential and external-location tools let you script the entire Unity Catalog setup in a single Switchy session, referencing your cloud IAM roles and storage paths without switching between the Databricks console and cloud provider UIs. This is the MCP's strongest use-case: one-time or quarterly infrastructure changes that require precise configuration across multiple cloud accounts. If you're doing this setup more than once a quarter, write a Terraform module instead—but for ad-hoc multi-cloud onboarding, the MCP is faster than clicking through the Databricks UI.
Frequently asked
What does the Databricks MCP let me do in Switchy?
It connects your team's Databricks workspace so AI assistants can manage Unity Catalog objects, run and cancel jobs, assign metastores, create external locations, and handle security group membership. You're essentially giving Switchy the ability to orchestrate your data platform setup and governance tasks without leaving the chat interface.
Do I need workspace admin rights to connect Databricks?
Yes, for most operations. Creating metastores, assigning them to workspaces, managing external locations, and modifying security groups all require account-level or workspace admin permissions in Databricks. OAuth2 will prompt for the scopes your team's role allows, but if you're not an admin, many of the 50 tools will return permission errors.
Can the MCP query data from my Databricks tables?
No. This integration manages Unity Catalog infrastructure and job orchestration—it doesn't run SQL queries or read table contents. If you need to query data, use Databricks SQL directly or connect a BI tool. The MCP is for provisioning catalogs, credentials, and external locations, not for analytics workloads.
Why use this instead of the Databricks CLI or REST API?
The MCP wraps Databricks APIs into natural-language tools that AI assistants understand. Instead of writing curl commands or Python scripts to create a metastore or cancel a job, you describe what you want in Switchy and the assistant calls the right tool. It's faster for ad-hoc admin tasks but doesn't replace automation pipelines.
Who on my team should connect the Databricks integration?
Whoever owns your Unity Catalog setup and workspace configuration—typically a data platform engineer or admin. Regular data scientists and analysts won't have the permissions most tools require. One shared connection per workspace is usually enough; individual users don't need separate OAuth flows unless you're isolating permissions across projects.