> For the complete documentation index, see [llms.txt](https://help.daasity.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.daasity.com/core-concepts/ai-analyst/connect-your-ai-mcp.md).
# Connect Your AI (MCP)
Connect the AI tool you already use - Claude and Cursor (+ others coming soon) — directly to your Daasity data. Ask questions in plain English and get answers from your own governed metrics: the same definitions that power your dashboards and Explores. No SQL, no exports, no dashboard-building.
{% hint style="success" icon="lightbulb" %}
**How this works:** You set the connection up once inside Daasity (workspace menu → **MCP**), but you do your actual asking inside your own AI tool. Daasity is where you turn it on; your AI tool is where you use it. This is separate from the in-app AI Analyst assistant, which you use inside Daasity itself.
{% endhint %}
### Why it's valuable
**Answers come from your governed metrics - not a guess.** Your AI queries Daasity's semantic layer, so "net sales," "orders," and "new vs. returning customers" mean exactly what they mean everywhere else in Daasity. The numbers tie back to your dashboards.
**No SQL, no exports.** Ask in plain language. You don't need to know table names or download a CSV to give your AI something to work with.
**One place to think.** Blend a data question into the same AI conversation where you're already planning, writing, or analyzing.
**Secure by design.** Every connection is read-only, scoped to your workspace only, and revocable at any time.
#### **Why the connection uses your semantic layer, not raw tables**
You ask in plain English, but your AI still writes a query behind the scenes to fetch the answer - the difference is *what* it queries. A raw database connection hands the AI raw tables and leaves it to guess the joins, filters, and metric definitions, which is exactly how you get numbers that don't match your reports. Daasity's connector points that same query at your governed semantic layer instead. The metrics, joins, and filters are already defined and validated in your Daasity models - so the query your AI writes resolves against definitions you can trust, not raw tables it has to interpret.
**Your Daasity configuration is what makes this work.** This isn't a separate AI that magically knows your business - it reads the integrations, extractors, workflows, and modeled logic you've set up in Daasity. If something isn't configured and modeled in your account, the AI can't answer questions about it. A clean, complete setup produces accurate, complete answers.
Before you start
You'll need:
* A Daasity workspace with **AI Analyst** enabled.
* An AI tool that supports MCP connectors - today **Claude Desktop**, **Claude on the web (claude.ai)**, and **Cursor**. Support for **ChatGPT** and others follows as they ship MCP.
* Permission to add a connector in your AI tool. On personal Claude plans (Free/Pro/Max) any user can. On **Team/Enterprise Claude accounts, an Owner must add the Daasity connector to the organization once first** (see Troubleshooting below).
### Connect your AI (recommended)
#### This path lets your AI tool handle sign-in for you - no tokens to copy or store.
{% stepper %}
{% step %}
### **Open the MCP screen in Daasity.**
Click your workspace name (bottom-left) → **MCP**. This opens the **Connect Your AI** screen.
{% endstep %}
{% step %}
### **Copy your MCP server URL.**
On the **Connect via AI tool** tab, pick your AI tool and click **Copy** next to the URL. Exact URL will be shared in-app. Always copy the URL from your own screen - it's specific to your environment.
{% endstep %}
{% step %}
### **Add the connector in your AI tool.**
In Claude Desktop: Settings → **Connectors** → **Add custom connector** → paste the URL → **approve** in the browser tab that opens. The steps are nearly identical for Claude on the web and Cursor; the Daasity screen shows the exact steps for whichever tool you pick.
{% endstep %}
{% step %}
### **Start asking.**
When you approve, Daasity issues a connection to your workspace. Go back to your AI tool and ask a question about your data.
{% endstep %}
{% endstepper %}
#### **Alternative: manual token**
If your tool doesn't support the guided flow, open the **Manual token** tab, choose your AI, click **Generate Token**, and paste the token into your tool's MCP configuration. Treat this token like a password - anyone with it can query your data until it expires or you revoke it.
#### What you can ask
Because the connection uses your Daasity semantic layer, your AI can answer across the same subject areas you already analyze in Daasity:
* **Sales & products:** "Top 10 products by net sales last 30 days" · "Revenue by channel last 90 days" · "Total omnichannel dollar sales by retailer last quarter"
* **Customers & retention:** "New vs. returning revenue this month" · "LTV by first-order channel" · "Active subscribers and who's at risk of churning"
* **Marketing:** "ROAS and CPO by channel and vendor last 30 days" · "First-click vs. last-click attribution"
* **Operations:** "Which products are lowest on weeks of supply by store?"
Beyond one-off questions, teams use the connection to:
* **Retire the report-request queue.** Anyone on the team can self-serve a quick number, ranking, or trend instead of filing a request and waiting.
* **Prototype and explore on the fly.** Build an interactive view in chat (e.g., LTV vs. AOV by cohort), refine it conversationally, and use your Daasity dashboards as the QA reference for anything that graduates to a scheduled report.
{% hint style="info" %}
**Retrieval vs. analysis:** pulling numbers is fast and reliable. Asking for judgment or a formal analysis depends on how you prompt it - keep questions open-ended and keep a person in the loop to sanity-check before handing results to others.
{% endhint %}
#### Managing connections
The **Connect Your AI** screen lists every connection under **Your connections** with its status, when it was added, last used, and when it expires.
* **Revoke access:** click **Disconnect** next to a connection - the AI tool loses access immediately.
* **Renew:** reconnect using the steps above. Expired connections don't renew themselves (intentional, for security).
* **Multiple connections are fine:** add several tools or people and revoke any one independently.
* **Keep your AI's view current:** definitions refresh automatically when your data model changes. To force a refresh, use **Refresh MCP Views → Refresh Now**.
#### Security & governance
* **Read-only** - can query data, never change, delete, or write it.
* **Tenant-isolated** - a connection only ever sees your workspace's data.
* **Time-limited** - your connection is time-boxed and must be periodically re-approved; the underlying access token is refreshed automatically in the background.
* **Usage-limited** - queries draw against a workspace token quota and a per-user daily query quota, so heavy or runaway usage is capped automatically.
* **Revocable** - disconnect any connection at any time.
* **Governed** - answers use your validated Daasity semantic layer. Queries run under your workspace's governed data access - all users on a connection query the same modeled, permissioned dataset.
Troubleshooting & FAQ
**My AI tool says the connection failed.** Re-copy the MCP server URL from the Daasity screen (it must be exact), confirm you completed the approval step in the browser, and make sure your tool supports custom MCP connectors.
**My AI can't find a metric or field I know exists.** Either it isn't modeled in your Daasity data (the connector only surfaces what your configuration contains - confirm the integration/metric is set up), or your model changed recently and the AI hasn't picked it up (**Refresh MCP Views → Refresh Now**).
**Does the AI just know all my data automatically?** No. It reads what you've configured and modeled in Daasity. If it isn't set up in your account, the AI can't answer about it - your configuration is what makes answers accurate and complete.
**The connection stopped working after about a month.** Connections are time-boxed and expire periodically by design. Reconnect using the steps above.
**I'm on a Team/Enterprise Claude account and there's no "Add custom connector" option.** Expected - it's a one-time admin step, not a bug. On personal plans any user can add a connector; on Team/Enterprise, only an Owner/Primary Owner can add one to the organization. Once they add the Daasity URL once (Claude → **Organization settings → Connectors → Add**), everyone else sees it under a **"Custom"** label and connects individually with their own credentials.
**Can I connect more than one tool or person?** Yes - each connection is listed and revoked independently.
**Which tools are supported?** Claude Desktop, Claude on the web, and Cursor today; ChatGPT and other MCP clients as they add support. Any MCP-compatible client can use the manual token method.
**I don't see the MCP option at all.** AI Analyst may not be enabled for your workspace, or it's locked by SSO and controlled by an admin. Contact your Daasity account team.
**My connection shows "Ready" with a valid expiration date, but questions fail with an authentication/token error.** This can happen because your AI tool's connection to Daasity and the underlying warehouse access token are managed separately - the connection can show as active while the token needed to actually run a query has lapsed. Disconnect and reconnect the affected connection (**Your connections → Disconnect**, then reconnect using the steps above) to force a fresh token. If this recurs frequently, contact your Daasity account team.