Berner Setterwall6 min read

    Snowplow Is Live in Cogny: Your Tracking Setup, One Prompt Away

    Snowplow is behavioral data done properly: self-describing event schemas, a tracking plan your whole team agrees on, and a warehouse you own. But the setup lives in the Snowplow Console, separate from the data, with its own API keys and its own vocabulary. Now your agent can reach it: Snowplow is an MCP integration in Cogny, and your agent can read and work on your tracking setup without ever seeing raw events.

    No warehouse connection and no SQL: one scoped API key created in the Console, pasted once into Settings → Cogny Managed MCPs, and your agent is in.


    TL;DR

    • Snowplow is live now. Connect it with a scoped API key from your workspace Settings → Cogny Managed MCPs.
    • The tools cover the pieces of a Snowplow tracking setup: data structures (schemas) and their drafts, tracking plans, event specifications, pipelines and the data catalog.
    • Tracking-setup reads. List and fetch schemas and their versions, tracking plans, event specs and pipelines, and search the data catalog, so an agent refers to the right event or entity without guessing.
    • Create tools add new data products, event specs and schema drafts; update tools overwrite an existing data product or event spec, published ones included, and ask for confirmation first. Deploying a schema to a pipeline environment asks for confirmation too.
    • Raw event data stays in your warehouse. This connector covers the tracking setup; the events themselves are reached through a database connection.
    • Works from any MCP client on app.cogny.com/mcp.
    • Included in Cogny Solo ($9/mo) and every plan above it; on Solo your own agent's standard MCP tool calls run under fair use.

    What you can do with it

    AreaWhat your agent can do
    Data structuresList the schemas in your organization and where each is deployed; get one, or a specific version's JSON schema; list and read schema drafts; validate a schema; create a draft of a new one
    Tracking plansList and read your tracking plans, optionally with event volumes; create a new one; change an existing one's name, description, domain, owner or source apps
    Event specificationsList the events in your tracking plans (filter by plan, source schema or status); read one with its compatibility warnings; create one; edit, publish or deprecate one
    PipelinesList your production and development pipelines and read one's status and configuration
    Data catalogList the events and entities in your setup and which are tracked together, or search them by name or description

    In plain terms: "What schemas do we have for checkout events, and which environment is version 1-0-1 deployed to?" "Create a draft tracking plan for the mobile app and add an event spec for the login event." "Which events carry the product entity?" The answers come from Snowplow's own API, against your live tracking setup.

    One boundary: setup, not raw data

    Snowplow's Console API is about the tracking: the schemas, tracking plans and pipelines that shape how events arrive. The raw event data lives in your own warehouse (BigQuery, Snowflake, Redshift or Databricks), and this connector stops at the Console. When an agent needs the events themselves (drop-off between two event types, event volume per day), that's a database connection, reached the way every other Cogny data connection works. The two fit together: the connector keeps the tracking plan legible, and the warehouse connection answers questions about the events.

    New things run; changes to live things wait for you

    Creating something new runs straight away: a schema draft, a schema validation, a new tracking plan, a new draft event spec. None of them changes anything already in use.

    Changing something that's already in use waits for you. update_data_product and update_event_spec overwrite an existing tracking plan or event spec, published ones included, so the first call returns the exact change it would make and nothing is sent. Your agent shows it to you, and the change goes through only when it re-runs the tool with confirm=True. The same applies to deploy_data_structure, which ships a schema version from one environment to another (validated to development, development to production), and to creating an event spec that's already published.

    Updates keep everything you didn't ask to change. Rename a tracking plan and its description, owner, domain and source apps stay as they were.

    The API key decides what the agent can touch

    The connector follows the key's permissions. Create the Snowplow API key with View on the features you want to read and every read tool works. Add Manage on the features you want your agent to change and the write tools work there too. Deploying to production also follows Snowplow's own rule: only keys with admin rights can do it. Use a dedicated scoped key for Cogny rather than a legacy Global admin key, and delete it in the Console if you ever disconnect.

    Two ways in: chat or scheduled

    Ask from any MCP client, or let a scheduled report review your tracking setup on a cadence. The Data & Tracking Quality report reads your Snowplow tracking plans, event specs and schema deployments when Snowplow is connected, and flags drafts that never shipped and specs with schema warnings. Reports read; they don't change your tracking plan.

    How to connect it

    1. In the Snowplow Console, open Settings → Manage organization and copy your Organization ID.
    2. Under View and manage API keys, click Create API key and choose the type "Snowplow API key". Grant View on the features you want to read, and Manage on any you want your agent to change. Copy the API Key ID and the API Key; the key is shown only once.
    3. In your workspace Settings → Cogny Managed MCPs, click Connect on the Snowplow card.
    4. Paste the Organization ID, API Key ID and API Key.

    Disconnect any time with one click, and delete the key in the Console if you stop using it.

    On trust

    Every tool says what it does to MCP clients: reads are marked read-only, and the tools that overwrite a tracking plan or event spec or deploy a schema are marked as changes that are hard to undo. Those never run without your confirmation, and anything new your agent creates starts as a draft unless you ask otherwise.


    If your team runs its behavioral data on Snowplow, or you're getting a tracking plan under control, this puts your schemas, tracking plans and pipelines one prompt away. Connect it from your Cogny dashboard and ask your first question.