How to Connect Databricks to Claude (and What It Can't Do)
Databricks ships its own managed MCP servers, so Claude can run natural-language queries against your lakehouse — but only inside a chat you start, and only over data you point it at. Add a Databricks managed MCP endpoint as a custom connector in Claude, authorize through Unity Catalog, and Claude can turn “what were weekly active users by plan last quarter?” into Genie SQL and hand you the answer. This is one of the more genuinely useful — and more dev-leaning — Claude connections, so it’s worth being precise about what it does and where it stops.
Below: what the managed servers expose, how the connection is governed, why “ask a question” is not the same as “watch a metric,” and how to put a job or query result on an actual trigger.
What Databricks’ managed MCP servers expose
Unlike a community wrapper, these come from Databricks itself. The managed MCP servers are hosted by Databricks, authenticate through Unity Catalog, and cover three areas:
- Genie spaces — natural-language-to-SQL. Claude sends a question, Genie translates it against the tables in a given Genie space and returns rows. This is the headline capability: conversational analytics over Delta tables.
- Unity Catalog functions — call registered SQL/Python functions as tools, so Claude runs governed, pre-defined logic rather than arbitrary queries.
- Vector Search — semantic search across indexed tables, useful for retrieval over docs, tickets, or product data sitting in the lakehouse.
Because everything routes through Unity Catalog and the AI Gateway, Claude only ever sees what the authorizing principal is granted. In practice you type things like:
- “Using the revenue Genie space, show gross margin by region for the last four quarters.”
- “How many Stripe payment failures did we log yesterday versus the trailing 30-day average?”
- “Semantic-search the support-ticket index for anything mentioning SSO timeouts this week.”
Databricks also documents MCP on the agent framework side, so the same servers that answer your Claude questions can back a production agent later.
Connecting it is a paid-plan, custom-connector job
Databricks isn’t a one-click app in Claude’s connector directory — it’s an MCP endpoint you add yourself:
- In your Databricks workspace, get the managed MCP server URL for the Genie space (or Unity Catalog / Vector Search) you want to expose.
- In Claude, open Settings → Connectors and choose Add custom connector.
- Paste the server URL and complete the OAuth flow — Databricks authenticates you and Unity Catalog scopes what Claude can touch.
- Ask a question against the Genie space to confirm it’s live.
Two gates worth flagging up front: custom connectors require a paid Claude plan, and the useful surface requires someone with Databricks and Unity Catalog admin rights to set up the Genie space and grant the right permissions. This is not a five-minute consumer setup; it’s a platform integration.
Why “ask a question” isn’t “watch a metric”
The Databricks MCP servers are, sensibly, built for querying — and querying is a pull, not a push.
Nothing fires when a number moves. Claude will happily tell you failed payments spiked when you ask. It has no way to notice the spike on its own, because Claude connectors carry no event triggers and no schedules. A job that finishes at 2am, a query result that crosses a threshold at noon — Claude learns about either the next time you open a chat and type the question.
It reports; it doesn’t route. Claude can read the answer out of Genie. It can’t then open a Jira ticket, page the on-call engineer, or drop the number into the exec channel. Those are write actions in other systems, outside what a read-oriented analytics connector is for.
The permission model is a feature, not a bug — but it’s yours to run. Unity Catalog governance is exactly what you want for a lakehouse. It also means the value depends entirely on how the Genie space and grants are configured. Point Claude at a badly modeled space and you get confident, wrong SQL.
So Claude is a strong desk-side analyst for “what does the data say right now?” — and structurally unable to be the thing that does something the moment the data changes.
Data that reacts on its own: Carly
Answering a question is half the job; reacting to the answer is the other half. A job failure should page someone. A metric crossing a threshold should open a ticket and post to Slack — at 2am, with your laptop shut. Carly — an AI executive assistant that runs on triggers, in the cloud — handles that half:
- A Databricks job’s result crosses a threshold (say, payment-failure rate above 2%) → Carly creates a Jira ticket, posts the number and a chart link to the finance channel in Slack, and emails the on-call lead. No chat opened.
- A nightly pipeline finishes → Carly pulls the summary row, formats a plain-English digest, and sends it to the Stripe reconciliation team every morning at 7am on a schedule.
- You describe it in plain English — “when the churn query comes back above target, open a ticket and brief me” — and Carly interviews you, then builds the workflow with you. Nothing to host beyond the connection.
AI agents start at $35/month, and workflow steps that don’t use AI — the ticket creation, the Slack post, the scheduled send — run free and unlimited. Databricks is one of 200+ tools Carly connects to; see the Databricks integration page and the full integrations list.
To be honest about fit: this is a more technical integration on both sides. If your team already lives in Genie and just wants conversational SQL, the Databricks MCP server is excellent on its own. Carly earns its place the moment you want a result to trigger an action.
Side by side
| Claude + Databricks MCP | Carly | |
|---|---|---|
| Natural-language queries via Genie | Yes | Yes (or reads Genie output) |
| Unity Catalog governance | Yes | Yes (inherits your grants) |
| Reacts when a job finishes or a metric moves | No | Yes, on triggers |
| Opens a ticket / pages on-call / posts to Slack | No | Yes |
| Runs a scheduled morning data digest | No | Yes (cloud) |
| Setup | Custom connector + paid Claude plan + UC admin | Plain-English interview |
| Pricing | Paid Claude plan (MCP free with Databricks) | AI agents from $35/mo |
Frequently Asked Questions
Does Claude integrate with Databricks?
Yes. Databricks publishes managed MCP servers for Genie spaces, Unity Catalog functions, and Vector Search. Add the server URL to Claude as a custom connector (paid plan required) and authorize through Unity Catalog. Claude can then run natural-language queries and semantic search over the data you’ve granted.
Can Claude run SQL on my lakehouse through the connector?
Effectively, yes — via Genie. Claude sends a natural-language question, Genie translates it to SQL against the tables in that Genie space, and returns the rows. It’s read/query oriented; Unity Catalog scopes exactly which tables and functions Claude can reach.
Will Claude alert me when a Databricks job or metric changes?
No. Claude connectors have no event triggers or schedules, so Claude only reports on a job or metric when you open a chat and ask. For threshold-based alerts, tickets, and scheduled digests, use a trigger-based agent like Carly.
Is the Databricks Claude connection a one-click app?
No. Databricks isn’t in Claude’s one-click connector directory as of mid-2026 — you add the managed MCP server URL yourself under Settings → Connectors, and someone with Databricks and Unity Catalog admin rights configures the Genie space and grants.
What does automated Databricks alerting cost with Carly?
AI agents start at $35/month, and the non-AI steps — creating the ticket, posting to Slack, sending the scheduled digest — run free and unlimited, so a result-to-ticket-to-brief pipeline stays affordable even at query volume.
More: Claude connectors · Can Claude send emails · Claude vs Carly · Claude Cowork alternatives · Claude + Datadog · Claude + Sentry · Claude + Airtable · Best AI agents for productivity
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
