For AI agents: a Markdown version of this page is at https://docs.snowplow.io/docs/signals/applications.md. The index of all documentation pages is at https://docs.snowplow.io/llms.txt, and the index for this section is at https://docs.snowplow.io/docs/signals/llms.txt.
Choose how to use Signals attributes
Once you've defined and published attribute groups, several kinds of consumer can read what Signals calculates. Applications can pull values when they need them, or Signals can push an intervention when a user meets your criteria. Pull and push compares the two patterns.
Use this table to find the page for what reads the data:
| What reads it | What it gets | Use | Interfaces |
|---|---|---|---|
| Application code | Calculated attribute values for a user or session | Services, then retrieve attributes. Services are recommended for production. | Python SDK, Node.js SDK, Signals API |
| An agent that needs recent session activity and a prompt | The session's latest events, up to 100 events or 1 hour, as JSON or a narrative | Agentic contexts, scoped to domain_sessionid | Python SDK, Node.js SDK, Signals MCP server |
| An agent that chooses what to read at runtime | Published attribute groups and agentic contexts, as tools | Signals MCP server | Any MCP client |
| An application that reacts when criteria are met | An intervention payload, pushed as soon as the criteria are met | Define interventions, then subscribe to them | Python SDK, browser plugin, Signals API |
| Offline model training | Point-in-time attribute values for labeled anchors, in a warehouse table | Build training datasets | Python SDK |
Agents
An agent can read Signals data in two ways. To ground it in what the user is doing in the current session, retrieve an agentic context and pass it to the model. To let the agent list and read published attribute groups and agentic contexts itself, add the Signals MCP server to its tools.
For context about a logged-in user across sessions, use an attribute group keyed on user_id or domain_userid; see identity resolution.