One workspace, every step of the pipeline.
A governed layer, built to be read by machines.
Every Agent View sits in the control plane: the part of your stack built for agent consumption instead of BI dashboards. Governance is not bolted on afterward — it travels with the data itself, in every response an agent gets back.
Definitions travel with the data
Column meanings, business logic, and units are embedded in the view itself, not left in a wiki an agent can't reach.
Access enforced at the view
Row- and column-level permissions apply before an agent ever sees a record. What a user can't see, their agent can't either.
Lineage attached
Every field traces back to its source table and transformation, so an agent's answer can point to where the number came from.
Versioned and documented
Changes are tracked over time. Agents and humans always work from the same current source of truth.
Everything an agent needs, in one place.
An Agent View isn't just a SQL view with a friendlier name. It's a manifest: the query that builds it, the definitions that explain it, the policy that governs it, and the lineage that backs it — all reachable by the same MCP call.
Below is a simplified example of the revenue_360 Agent View that powers the demo on our homepage.
name: revenue_360
description: "Quarterly revenue by account, region, and stage"
source: fct_opportunity
definitions:
arr_delta: "change in annual recurring revenue, quarter over quarter"
access: role:revenue-team, row_filter:region
lineage: stg_salesforce → fct_opportunity → revenue_360
exposed_as: mcp.tool.query_agent_view("revenue_360")
Discover, query, and get an answer with receipts.
Agent lists Agent Views
Claude or Copilot calls the Dagen MCP server and sees which Agent Views it's authorized to reach.
Agent calls a scoped tool
The agent queries the Agent View directly. Definitions and access policy apply automatically, every time.
Response includes the receipt
The answer comes back tagged with the source table, the Agent View, and the governance under which it ran.
Everything you need to connect an agent to your data.
Quickstart
Add the Dagen MCP server and ask your first question against a working Agent View in about 10 minutes.
Read the quickstart ReferenceMCP reference
The tools and resources the Dagen MCP server exposes, with request and response examples.
Open the reference ConceptAgentic runtime
How the build, monitor, and heal agents work together — and how to set an autonomy level.
Learn the runtime SolutionConnectors & sources
Connect databases, SaaS apps, files, and streams without replatforming your existing stack.
See AI-ready data SupportTalk to an engineer
Questions about governance, deployment, or the runtime that the docs don't answer yet? Ask us directly.
Book a technical demoConnect Claude in 10 minutes.
One command to add the MCP server, one Agent View to point it at, one question to ask. That's the whole quickstart.
$ claude mcp add dagen https://app.dagen.ai/mcp
# then, in Claude
> what were EMEA bookings last week?
→ answered from agent view revenue_360
Where Agent Views get built and governed.
Build and grow with Dagen.
Technology partners extend the control plane. Consulting partners deliver it.
Technology partners
Warehouses, lakehouses, orchestrators, and BI tools that Dagen sits alongside — Snowflake, Databricks, BigQuery, dbt, Airflow, and more. Get listed, get connector coverage, and get in front of teams modernizing for agents.
Become a technology partnerConsulting & SI partners
Data and analytics consultancies delivering Agent Views and agentic pipelines to their own clients, with Dagen as the underlying platform. Implementation support and co-selling included.
Become a delivery partnerWorks with the stack your partners already run.
Agent Views, answered.
Is an Agent View just a database view?+
It's built on top of one — but a database view has no definitions, access policy, or lineage attached. An Agent View carries all three, so an agent can use it safely, no human translation required.
Does it replace my BI semantic layer?+
No. Your semantic layer stays put, serving dashboards and analysts. Agent Views are a parallel layer for a different consumer — agents asking thousands of ad hoc questions instead of loading a fixed report.
How is this different from a data catalog entry?+
A catalog entry describes data for a human to read. An Agent View is queryable — an agent calls it directly over MCP and gets rows back, governed the same way a person's access would be.
See an Agent View answer a live question.
A 30-minute technical demo, live against a working Agent View — bring your hardest question about governance.