business resources
11 Best Claude Connectors for Data Analysis in 2026, Ranked
16 Nov 2026

Sponsored content
Anthropic's Connectors Directory lists 1,128 connectors as of November 2026, and the data tools share that shelf with inbox search, calendars, and wikis (scrolling the whole thing counts as a light workout).
Here are the 11 best Claude connectors for data analysis, ranked by how much real analysis work each one lets Claude do on your company's data.
Short on time? Here's the shortlist:
Prophecy DataGPT is the best Claude connector for data analysis in 2026, because it does prep, reconciliation, and reporting inside Claude, shows a visual workflow and open code you can check, and reruns approved work on Snowflake, Databricks, or BigQuery.
Snowflake suits Cortex-ready warehouse teams, Google BigQuery is the free-to-try Google Cloud option, and Databricks answers through Genie. dbt brings lineage and metrics, Hex fits notebook teams, and Metabase covers open-source BI.
For product data, PostHog suits engineers, Amplitude handles growth experiments, and Mixpanel owns funnels, while Google Drive reaches the stray spreadsheets outside your warehouse.
| Connector | Best for | Reads or writes? | Vendor plan needed |
|---|---|---|---|
| Prophecy DataGPT | Governed prep and reporting | Reads and writes | Account, free to start |
| Snowflake | Cortex-ready teams | Set by Snowflake roles | Snowflake, plus admin setup |
| Google BigQuery | Google Cloud shops | Reads; execute_sql can write | Free sandbox works |
| Databricks | Lakehouse teams on Genie | Check vendor | Databricks with Genie One |
| dbt | Lineage and metrics | Reads, triggers jobs | Any dbt platform plan |
| Hex | Notebook-first analysts | Reads; edits need Editor role | Hex Team or Enterprise |
| Metabase | Open-source BI teams | Reads and writes | Any Metabase plan |
| PostHog | Product engineers | Reads and writes | Free to connect |
| Amplitude | Growth and experiments | Reads and writes | Check vendor |
| Mixpanel | Funnels and retention | Check vendor | Free plan and up |
| Google Drive | Files and exports | Reads and writes | None stated |
What does a Claude connector do for data analysis?
A connector plugs Claude into an outside service through MCP, the Model Context Protocol, an open standard that lets AI apps talk to other tools.
Once connected, Claude can pull your data and take actions in that service, and Anthropic says it inherits each person's permissions from the connected service.
The vetted ones live in the Connectors Directory, and directory connectors work on every Claude plan, Free included. Anything else is a custom connector pointed at a remote MCP server, and Free users get one of those (choose it the way you'd pick a desert-island album).
On Team and Enterprise plans, an Owner or Primary Owner has to enable each connector for the organization before members can use it, and each person still signs in individually. Owners can also let a connector read from a service while blocking it from writing anything back.
Scope varies a lot between connectors. ClickHouse's connector is read-only by design, BigQuery's has one tool that can change data and caps results at 3,000 rows, and PostHog's can delete insights and feature flags.
Then there are plugins, which bundle skills, connectors, and sub-agents into one package on paid plans. The Claude Data plugin from Anthropic writes SQL, builds charts and dashboards, and works from pasted results or uploaded CSV and Excel files when no warehouse is connected.
Claude for Excel is another thing again: an add-in on paid plans that works on the spreadsheet in front of you and can pull context from connectors in its sidebar. Neither one is a connector, so neither gets a numbered spot on this list.
How I picked these connectors
Where the data lives. A connector helps only if it reaches the system your numbers sit in, whether that's a warehouse, a product analytics tool, or a folder of exports.
What Claude can do. Reading, querying, and writing are three different powers, and I gave write access credit only where the vendor documents controls around it.
Who sets it up. Some connectors work after a sign-in, while others need a Snowflake admin or a curated Genie setup before anyone types a question.
What the vendor side costs. Directory connectors work on every Claude plan, but the service on the other end often has its own price, so each pick lists the plan you need.
1. Prophecy DataGPT: best overall for governed, repeatable analysis
If your analysis has to survive a second look from finance, this is the connector to install first.
Anthropic lists it as verified in the directory's Data and analytics category, where its tagline reads: governed data work for business teams, right in Claude. You ask for a reconciliation or report in plain language, and specialist agents handle the cleaning, joins, and calculations.
Results appear in the chat as interactive views, like a pipeline canvas, a data table, or a dashboard. Click any step to see its inputs, logic, data, and code, and if a join looks wrong, say so in plain English and the workflow updates.
The app keeps the rules you confirm, like a fiscal calendar or a vendor mapping, so the next request from anyone on the team starts from them. Approved workflows rerun on a schedule on Snowflake, Databricks, or BigQuery without prompting the model again.
Only metadata goes to the LLM provider, and your existing access controls apply. Uploaded files skip the chat history, so a 40-tab workbook works as well as a small CSV.
Best for: finance, tax, marketing, and operations analysts who need reconciliations or recurring reports they can defend and rerun.
What Claude can do with it: read and write. Its 54 tools create, read, update, and delete data assets and run and edit pipelines, and you grant all permissions together. On Team and Enterprise, an Owner adds it once and each member signs in with their own account.
What it costs: you can start free, with 20 credits a month running on embedded DuckDB. Professional, at $150 per user per month, moves processing into your Databricks, Snowflake, or BigQuery, and Enterprise Express runs $4,000 a month on annual billing for teams of up to 20.
Watch out for: you need a separate account with the app alongside Claude, which is free to start, and heavier use runs on its credits.
2. Snowflake: best for teams already running Cortex
Snowflake's connector is the strongest warehouse pick for teams that already built Cortex Agents and semantic views and want Claude to put questions to them.
Snowflake makes it, the listing sits in the Data and analytics category, and its tools are Cortex Agents and Cortex Search. Snowflake says you can use natural language to execute SQL queries, which puts warehouse answers within reach of business users.
The setup is where the work hides, because a Snowflake admin has to create the MCP server and grant privileges on it first. Snowflake's docs add that access to the server doesn't give access to the tools, so each tool needs its own grant.
Snowflake also recommends exposing a Cortex Agent as the only tool Claude sees and keeping direct SQL behind a separate MCP server with a least-privileged role. That's sound advice, and it means your first week with this connector is mostly about permissions.
Best for: teams already on Snowflake with Cortex semantic views in place who want governed answers in chat.
What Claude can do with it: ask Cortex Agents questions, search with Cortex Search, and run SQL wherever your Snowflake role and tool grants allow it.
What it costs: Snowflake publishes no MCP-specific price. As of November 2026, credits on AWS US East list at $2.00 for Standard, $3.00 for Enterprise, and $4.00 for Business Critical, and AI features bill separately in AI Credits.
Watch out for: it isn't plug-and-play. Someone with Snowflake admin rights has to build and grant the server before your analysts see a single answer, so book that time before the rollout.
3. Google BigQuery: best for Google Cloud shops and free trials
Google BigQuery has the friendliest on-ramp of any warehouse connector on this list, because you can try it without turning on billing.
Google built it, it joined the directory in January 2026, and its tools read like a tidy checklist: list_dataset_ids, get_dataset_info, list_table_ids, get_table_info, and execute_sql. That makes it easy to explain to a security reviewer, since you can see every move Claude is allowed to make.
Google's docs say execute_sql is the only tool that isn't read-only, and you can restrict it with a deny policy if you want Claude to look and never touch. Queries stop at three minutes, and results top out at 3,000 rows.
That row cap shapes how you ask. Aggregations and summaries come back fine, so push the heavy lifting into the query and let Claude read the answer.
Best for: Google Cloud shops, and anyone who wants to test a warehouse connector before the budget conversation starts.
What Claude can do with it: browse datasets and tables, read schema details, and run SQL through execute_sql, which you can lock down with a deny policy.
What it costs: Google's setup steps work without billing or a credit card through the BigQuery sandbox. Users need the BigQuery Job User and BigQuery Data Viewer IAM roles.
Watch out for: the 3,000-row cap and the three-minute query limit. A question that needs row-level detail across a big table belongs in a scheduled job somewhere else.
4. Databricks: best for lakehouse teams that trust Genie
Databricks is the right call for lakehouse teams whose data team already curates Genie Agents, because the connector hands Claude's questions to Genie and lets it do the answering.
The listing is called Databricks Genie, and Databricks added it in November 2025. Claude sends a plain-language question, Genie searches your structured and unstructured data, writes SQL, and returns an answer, and Databricks says Unity Catalog permissions are always enforced.
Its tools mirror that back-and-forth: genie_ask, genie_poll_response, genie_get_query_result, genie_cancel_response, and view_ask. Claude asks, waits, fetches, and can cancel a question that's taking its sweet time.
Answer quality depends on the Genie Agents behind it, which Databricks called Genie spaces until July 2026. Those are domain setups where your data team defines trusted tables, metrics, and business rules, so well-curated agents give sharp answers and neglected ones give vague ones.
Best for: lakehouse teams that already trust their Genie Agents and want those answers available in Claude.
What Claude can do with it: ask Genie questions, poll for responses, fetch query results, and cancel requests.
What it costs: Databricks says the Genie One MCP server uses Chat in Genie One. Genie One is free for LLM usage until January 31, 2027, while compute bills separately and SQL Serverless lists at $0.70 per DBU.
Watch out for: Databricks says its Beta Genie endpoint reaches end of life on November 30, 2026, so check that your setup points at the current one.
5. dbt: best for lineage and governed metrics
dbt's connector is the one analytics engineers will reach for, because it puts lineage and governed metrics one question away.
Added in June 2026, it carries 60 tools that browse models, sources, and lineage, query metrics from the Semantic Layer, and trigger jobs on the dbt platform. Ask what sits downstream of a model you're about to change, and Claude can walk the lineage to answer.
Metrics come from the Semantic Layer, so revenue means the same thing in chat that it means in your reports (a small miracle in most companies). That alone makes it worth a trial for any team that has argued about a KPI definition this quarter.
dbt describes the server as a read-only access layer for data storage, but it can still trigger jobs. Treat it as a tool with real write-ish power and set approvals to match.
Best for: analytics engineers who want lineage, model metadata, and Semantic Layer metrics in chat.
What Claude can do with it: read models, sources, and lineage, query metrics, and trigger jobs on the dbt platform.
What it costs: dbt says the remote MCP server is on all dbt platform plans, though the APIs behind its tools vary by plan. A single developer seat on dbt is free, and Starter is $100 per user per month.
Watch out for: the remote server supports Semantic Layer, SQL, Discovery, Administrative API, and dbt v2 tools only, so CLI and codegen tools need the self-hosted version.
6. Hex: best for analysis that already lives in notebooks
Hex earns its spot if your team's analysis already lives in Hex projects, since the connector lets Claude find that work and talk to the Hex agent about it.
Anthropic files it under Productivity, which makes it easy to miss when you browse the data shelf. Its four tools are search_projects, create_thread, get_thread, and continue_thread, so Claude can locate an existing project and then start or continue a thread with the Hex agent.
That's a handy split of labor. Claude handles the conversation, and the notebook that your analysts already trust does the SQL and Python.
When you trial it, ask a question you already know the answer to and check whether the thread lands on the right project. That tells you more about your project naming than about the connector, which is useful too.
Best for: teams whose SQL and Python analysis already sits in Hex and who want to ask about it from Claude.
What Claude can do with it: search projects and run threaded questions through the Hex agent. Hex says users need the Explorer role or higher, and project editing tools need Editor or higher.
What it costs: Hex says its MCP server is available on the Team and Enterprise plans. Team lists at $75 per editor per month, and Hex offers a 14-day Team trial with no payment card.
Watch out for: you need that paid Hex workspace, so this pick only makes sense for teams that already run on Hex.
7. Metabase: best for open-source BI teams
Metabase is the pick for teams running open-source BI that want Claude to work against the same tables, questions, and dashboards everyone already uses.
Added in May 2026, it lets Claude search tables and metrics, explore field values, and construct or run queries. Tools include execute_query, create_question, and create_dashboard, so Claude can leave something behind in Metabase for the rest of the team.
That last part is the appeal. An answer that lands as a saved question gets reused, while an answer stuck in a chat thread gets asked for again next week.
Scopes are granular, too. One of them, agent:sql:run, lets Claude write and run raw SQL, and Metabase says a scope never grants more than your Metabase permissions allow.
Best for: open-source BI teams that want chat answers to land as saved questions and dashboards.
What Claude can do with it: read tables, metrics, and field values, run queries, and create questions and dashboards.
What it costs: Metabase says the MCP server is on every plan, while controls like MCP tools access and MCP analytics need Pro or Enterprise. Check with Metabase for plan prices.
Watch out for: the raw SQL scope. Hand it to the people who would write that SQL anyway, and keep everyone else on the search and question tools.
8. PostHog: best for product engineers
PostHog's connector is built for product engineers, and with 57 tools it's one of the busier listings in the directory.
Anthropic files it under Developer tools. It queries analytics with HogQL, PostHog's flavor of SQL, or natural language, builds and manages dashboards and insights, controls feature flags, and runs and monitors experiments.
That range is the draw and the risk. You can ask why signups dipped, build the insight that shows it, and check the experiment that might explain it, all from one chat.
The same chat can also delete an insight or a feature flag, because tools like insight-delete and delete-feature-flag ship with it. Set those to Needs approval in Claude unless you enjoy explaining to support why a feature vanished.
Best for: product engineers who want analytics, flags, and experiments in one conversation.
What Claude can do with it: read and write, from HogQL queries and dashboards to feature flags and experiments.
What it costs: PostHog says connecting to the MCP server and calling its tools is free, though some tools use LLMs internally and may be billed as PostHog AI spend.
Watch out for: write tools can change live flags, so review the per-tool settings before you connect a production project.
9. Amplitude: best for growth and experiment analysis
Amplitude fits growth teams that live in charts, cohorts, and experiment readouts.
Added November 2025 with 28 tools and separate US and EU connector URLs, it reads and writes your organization's data with your account permissions, according to Amplitude. Pick the URL that matches where your Amplitude data is hosted before you connect.
Write calls need the Use MCP (write) RBAC action, which gives admins a clean switch for who can change things from chat. Most growth teams will want a few people with that switch on and everyone else reading.
In a trial, ask Claude to compare two cohorts over the last quarter and then check the chart in Amplitude. If the numbers match, you've got a connector your growth team will use daily.
Best for: product and growth analytics, and experiment reviews.
What Claude can do with it: read and write within your account permissions, with writes gated by a dedicated RBAC action.
What it costs: check with Amplitude. The vendor says MCP service account authentication is available on the Enterprise plan.
Watch out for: Amplitude says the MCP isn't an ingestion endpoint, so it won't send new events into your project. Tracking still happens the usual way.
10. Mixpanel: best for funnel and retention questions
Mixpanel is the fastest route from "where do users drop off?" to a real funnel you can show your team.
Added in March 2026, it runs segmentation, funnel, and retention analyses, explores your event taxonomy, manages Lexicon metadata, and analyzes session replays. That last one is a nice touch, because a funnel tells you where people leave and a replay hints at why.
Lexicon access also makes it handy for cleanup work. Ask Claude which events have no description, and you'll learn a lot about how your tracking plan has aged.
Best for: funnel and retention questions, plus event taxonomy cleanup.
What Claude can do with it: run segmentation, funnel, and retention analyses, browse events, manage Lexicon metadata, and analyze session replays.
What it costs: Mixpanel says MCP is on by default for new free and growth accounts created after August 1, 2026. Free plans can be subject to Query API access limits.
Watch out for: older accounts need an organization admin to switch MCP on in the org's AI Settings first, so don't be surprised if the connection fails on day one. Lexicon edits change shared metadata, so decide who gets that power.
11. Google Drive: best for spreadsheets and exports outside the warehouse
Google Drive is the connector for the data nobody moved into a warehouse, which in most companies is a lot of data.
Google built it, files it under both Data and analytics and Productivity, and added it in March 2026. Claude can read and analyze Google Docs, Sheets, Slides, and PDFs, and upload files straight into your Drive.
It's the right tool for the vendor price list in a Sheet, the PDF statement from the bank, and the export someone dropped in a shared folder last quarter. Ask Claude to compare two of them, and you'll have an answer before you'd have found the files by hand.
Write access deserves a minute of thought, because the tools include create_file, update_file, share_file, and trash_file. Sharing and deleting are good candidates for an approval step.
Best for: spreadsheets, PDFs, and exports that live in Drive folders.
What Claude can do with it: read and write, including creating, updating, sharing, and trashing files.
What it costs: no vendor charge is stated on the listing.
Watch out for: it's a file connector, so analysis happens on file contents. Anything you need to rerun every month will outgrow it fast.
Which Claude connector should your team install first?
Start with where the work lands. If your analysts spend their week on reconciliations and recurring reports, install the top pick first, because a result you can check and rerun saves more hours than a faster query.
If your data team already curates a semantic layer, connect the warehouse that holds it next. Snowflake fits Cortex shops, Databricks fits teams with good Genie Agents, and BigQuery is the cheapest one to test.
Product teams should pick one of PostHog, Amplitude, or Mixpanel, whichever already holds your events, and skip the other two. Hex and Metabase make sense only if your team already uses them.
Whatever you pick, set each tool to Always allow, Needs approval, or Blocked before rollout, and add the Claude Data plugin on paid plans when people need SQL help across sources.
Where the best Claude connectors for data analysis go next
My bet is that the chat window becomes the place most analysis starts, for analysts and for the executives who used to email them. The connectors on this list already make that possible for reading data.
The open question is whether those answers can be checked and rerun. A number you can't trace gets rebuilt from scratch next month, so the connectors that win will be the ones that hand back their work and keep it running on a schedule.
FAQ
Are Claude connectors free to use?
Directory connectors are available on every Claude plan, including Free. The service on the other end may still charge, the way Hex requires a Team or Enterprise plan.
Can a Claude connector write to my database?
Many can, within your permissions in the connected service. Owners on Team and Enterprise plans can limit a connector to reading, and each tool can be set to Always allow, Needs approval, or Blocked.
What's the difference between a Claude connector and a Claude plugin?
A connector links Claude to one outside service, and a plugin is a bundle. A plugin packages skills, connectors, and sub-agents together on paid plans, and the Claude Data plugin is one example.






