AI agents are only as trustworthy as the context they work with. Today, we are excited to announce that the Collibra Model Context Protocol (MCP) server is available on Snowflake Marketplace and integrated with Snowflake’s public preview AI Gateway MCP & Tools app, making it easier than ever for Snowflake customers to bring governed business definitions, lineage, classifications, quality measures, semantics and data contracts directly into Snowflake CoWork and Snowflake CoCo.
This launch reflects a simple belief: governance should travel with your agents, right where the work happens. We will walk through exactly how to set up Collibra MCP in Snowflake and how your teams can start using it in their daily work.
Why Collibra MCP belongs in Snowflake
MCP is an open standard that gives AI assistants a live, structured connection to enterprise systems. The Collibra MCP server connects Snowflake agents to the governed metadata in Collibra, so an agent can answer critical governance questions before it acts:
- Is this table certified, or is it a staging copy or a deprecated feed?
- What does this business term mean, and which physical columns realize it?
- Where did this data come from, and what transformed it along the way?
- Who owns this asset, and which steward can I contact?
- Does this data contain sensitive information, and which policies apply?
- Can you share more information about the Data Product and its Data Quality scores?
- Update a metric, business description, or a policy from Snowflake and trigger a workflow in Collibra for stewards to approve
Collibra MCP can be used by business and technical users in Snowflake as well.
- For Business Analysts & Data Consumers:What is the official enterprise definition of 'Active Customer' for this report, and who is the designated data steward if I notice discrepancies?Is this sales dashboard using a certified data product or an unverified staging dataset?Which metric definitions apply to quarterly revenue calculation according to Collibra governance policies?
- What is the official enterprise definition of 'Active Customer' for this report, and who is the designated data steward if I notice discrepancies?
- Is this sales dashboard using a certified data product or an unverified staging dataset?
- Which metric definitions apply to quarterly revenue calculation according to Collibra governance policies?
- For Data Engineers & Developers:Which downstream data products or BI dashboards will be impacted if we modify the schema of this staging table?What source tables feed into this reporting model, and what transformations occur along the lineage pipeline?Are there any open data quality issues or failing quality checks registered in Collibra for this source table?
- Which downstream data products or BI dashboards will be impacted if we modify the schema of this staging table?
- What source tables feed into this reporting model, and what transformations occur along the lineage pipeline?
- Are there any open data quality issues or failing quality checks registered in Collibra for this source table?
- For Data Governance & Compliance Officers:Does this dataset contain PII or regulated metrics under GDPR/CCPA, and are the classification tags in Snowflake consistent with Collibra policies?Which business processes or third-party integrations currently use this sensitive data asset?Do Snowflake's automated classifiers match our registered classifications in Collibra, or do we need to trigger a reconciliation workflow?
- Does this dataset contain PII or regulated metrics under GDPR/CCPA, and are the classification tags in Snowflake consistent with Collibra policies?
- Which business processes or third-party integrations currently use this sensitive data asset?
- Do Snowflake's automated classifiers match our registered classifications in Collibra, or do we need to trigger a reconciliation workflow?
- For AI & Analytics Leaders:How can we ensure our custom AI agents only consume certified, high-quality data products without forcing users to manually double-check catalog entries?Can our AI assistant trigger a governance approval workflow in Collibra when updating data asset descriptions or metrics?How do we audit the governance context and data lineage leveraged by our AI models during execution?
- How can we ensure our custom AI agents only consume certified, high-quality data products without forcing users to manually double-check catalog entries?
- Can our AI assistant trigger a governance approval workflow in Collibra when updating data asset descriptions or metrics?
- How do we audit the governance context and data lineage leveraged by our AI models during execution?
Without this context, analysts and engineers switch tools, file requests with data stewards and wait. With Collibra MCP, the answers arrive in natural language inside the Snowflake experience they already use every day.
Before you begin
Make sure you have the following in place:
- A user who can register OAuth applications in your Collibra instance.
- A user with an ACCOUNTADMIN role. The new AI Gateway MCP & Tools flow is limited to ACCOUNTADMIN,
- Access to Snowflake CoWork, Snowflake CoCo or both for the users you plan to enable
Find Collibra MCP on Snowflake Marketplace
In Snowsight, open Snowflake Marketplace and search for Collibra MCP.
The Collibra listing provides an overview of the server, the tools it exposes and links to the setup documentation. The Collibra MCP server is free to use for Collibra customers.
Step 1: Register an OAuth application in Collibra
The connection uses OAuth, so every request runs under the identity of the person asking.
- In Collibra, go to Settings, then OAuth Applications
- Click Register user application
- Enter the Snowflake redirect URI: https://identity.snowflake.com/oauth2/callback
- Save the application, then copy the Client ID and Client Secret
Step 2: Add the Collibra MCP connector in Snowflake
The outline described below is the process as it functions today.
MCP connectors are now managed via Cortex AI Gateway
1.Open your AI Gateway MCP & Tools app
2. Search for the available Collibra MCP server
3. Enable it, enter your Collibra host, your registered application client id and secret.
4. Go ahead and decide who can use the MCP server.
5. Connect to load tools
6. Decide which tools to enable.
7. Find your new connection and add to your AI Client (optional)
For example, adding your Collibra MCP server to Cortex Code.
Follow the instructions on how to install your Collibra MCP server on your AI Client.
Step 3: Grant access to the right teams
.Grant access with SQL after enabling the MCP Server to the teams who should have access.
Step 4: Add Collibra MCP to a Cortex Agent, Snowflake CoWork & Snowflake CoCo
Snowflake Cortex Agent allows organizations to build custom, task-oriented AI assistants that automate complex analytical and data workflows directly within Snowflake. Snowflake CoWork serves as a collaborative workspace where business users, analysts, and data stewards can query, explore, and discuss data using plain language. Snowflake CoCo brings intelligent conversational assistance to everyday workflows, helping users discover insights and streamline tasks across the data platform. Adding Collibra MCP to this ecosystem enriches every interaction with governed business definitions, enterprise lineage, and data quality metrics, ensuring AI-driven insights and actions across these tools are accurate, trustworthy, and fully compliant.
To configure Collibra MCP with Snowflake Cortex Agents, Snowflake CoWork or Snowflake CoCo, you can check the official Snowflake documentation for more details.
Governance that travels with every request
Speed means little without trust, and this integration was designed with security at its core:
- Per user identity: Every tool call carries the authenticated user’s Collibra identity. If someone cannot see an asset in Collibra, they cannot see it through the agent either
- Server side credentials: Snowflake executes MCP tools on the server, and OAuth tokens never reach the browser
- Full observability: Snowflake records every Collibra tool call as part of the agent trace, giving you an auditable record of how and why an agent made a decision
- Instant control: Administrators can disable the integration at any time, which immediately revokes access for all users and agents
- Runtime tool discovery: New Collibra tools become available to your agents as soon as we publish them, with no connector updates required
- Built on an open standard: Collibra MCP connects to Snowflake through a standard MCP connector, so there is no custom engineering to build or maintain
See it in action
Now that you are set up, the next step is seeing what Collibra MCP can do in a real workflow. In our companion post, Governed AI at the speed of Snowflake: How Collibra MCP powers Snowflake CoWork and Cortex CoCo, we follow a quantitative risk analyst at a global bank who builds a BCBS 239 capital model entirely within Snowflake. Along the way, the agent verifies table certification, retrieves canonical glossary definitions, traces lineage across every transformation and updates the catalog, all without switching tools. It is a practical look at how the steps above translate into faster, audit ready work.
Find Collibra MCP on Snowflake Marketplace today, and give every agent in your organization the governed context it needs to act with confidence.
- Anand AkellaAnand AkellaSenior Partnership Account Manager Collibra
Anand Akella
Anand Akella
Senior Partnership Account Manager
Collibra
- Antonio CasteloAntonio CasteloHead of Partner Integration, Technology Partnerships, CollibraCollibra
Antonio Castelo
Antonio Castelo
Head of Partner Integration, Technology Partnerships, Collibra
Collibra
