How to use the Amplitude Automation via Rube agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file. The file holds instructions that an AI agent reads when a task matches the skill's description. This skill contains no code of its own.

The skill covers five kinds of Amplitude work: sending events, getting a user's activity, finding and identifying users, managing cohorts (saved groups of users), and browsing event categories. It also tells the agent to search for current tool schemas before acting, and to avoid common mistakes such as using the wrong user ID type or sending timestamps in seconds instead of milliseconds.

What you need before installing

The skill depends on an external service. You need a Rube MCP server connection configured in your agent client, using the endpoint https://rube.app/mcp. Adding that URL alone does not grant account access; the Rube service returns a connection and authorisation flow that you must complete.

You also need an authorised Amplitude account connected through Rube, and access to Composio's Amplitude toolkit, which is provided by the Rube service. The skill cannot work offline.

• Rube MCP server connection (https://rube.app/mcp) configured in your agent client
• An authorised Amplitude account connected through Rube via RUBE_MANAGE_CONNECTIONS
• Composio Amplitude toolkit access, provided by the Rube service

How to install the amplitude-automation skill

Download the ZIP for the skill from the page where you found this guide. Unzip it so that the amplitude-automation folder, containing SKILL.md, ends up in your personal skills folder at ~/.claude/skills/. To limit the skill to one project instead, place it in that project's .claude/skills/ folder.

If you use the Claude apps rather than a local setup, you can upload the ZIP instead. Open Settings, go to Capabilities, then Skills, and upload the file there.

• Personal install: ~/.claude/skills/amplitude-automation/
• Project install: .claude/skills/amplitude-automation/ inside your project

Connect Rube and Amplitude

Before running any workflow, the agent should confirm that Rube is reachable and that the Amplitude connection is active. The skill's setup sequence is shown below. If the connection is not ACTIVE, the returned auth link must be followed to complete Amplitude authentication.

Do not run any workflow until the connection status shows ACTIVE.

• Verify Rube MCP is available by confirming RUBE_SEARCH_TOOLS responds
• Call RUBE_MANAGE_CONNECTIONS with toolkit amplitude
• If the connection is not ACTIVE, follow the returned auth link to complete Amplitude authentication
• Confirm the connection status shows ACTIVE before running any workflows

How to trigger the skill with natural requests

You do not need special syntax. Describe the outcome you want in plain language and mention Amplitude, and the agent should match the request to the skill. Name the identifier you have, such as a user ID or email, and say whether you want to read, write or check something.

Examples of requests that fit the skill:

• Show me the last 50 events for the user with user ID 84213 in Amplitude.
• Find the Amplitude user with the email address jane@example.com and tell me their properties.
• Send a test page_view event to Amplitude for user ID test-001.
• List all my Amplitude cohorts and show the details of the one called Trial users.
• Add these five user IDs to the Churn risk cohort, then check whether the update has finished.

Understand the main workflows

Each workflow lists the tools the agent will call. Knowing them helps you judge what the agent is about to do.

Sending events uses AMPLITUDE_SEND_EVENTS. Getting user activity first calls AMPLITUDE_FIND_USER to resolve your ID, then AMPLITUDE_GET_USER_ACTIVITY. Identifying users uses AMPLITUDE_IDENTIFY with property operations. Cohort work uses AMPLITUDE_LIST_COHORTS, AMPLITUDE_GET_COHORT, AMPLITUDE_UPDATE_COHORT_MEMBERSHIP and AMPLITUDE_CHECK_COHORT_STATUS. Event categories use AMPLITUDE_GET_EVENT_CATEGORIES.

• Send events: each event needs an event_type and either a user_id or a device_id; time is in milliseconds
• Get user activity: needs Amplitude's internal user ID, which FIND_USER provides; results come newest first and are paginated with offset
• Identify users: property operations include $set, $setOnce, $add, $append and $unset
• Manage cohorts: only behavioral cohorts accept API membership updates; updates are asynchronous

Avoid the common mistakes

Most errors come from IDs and timestamps. Amplitude keeps its own internal user IDs, separate from your application's IDs, and activity lookup needs the internal one. Timestamps must be 13-digit millisecond values; a 10-digit seconds value will be read as a very old date. If you omit the timestamp, the server's receive time is used.

$set overwrites existing values, while $setOnce only writes if the property is not already set. Check these before running a write. Large cohort changes should be split into chunks, because membership changes per request may be limited.

• Use FIND_USER first to convert your user_id into Amplitude's internal ID
• Send times in milliseconds since epoch, not seconds
• Choose $setOnce instead of $set when you must not overwrite an existing value
• Chunk large cohort membership updates

Safety notes: scripts and network access

This skill runs no scripts. It does use the network, because every action is a request made through the Rube service to Amplitude. The skill sends and modifies real analytics and cohort data, so it is not suited to live production pipelines without human review.

Mistakes such as a wrong user ID, a $set overwriting values or a bulk cohort change can corrupt records. Review what the agent plans to write before confirming. Events and property changes are processed asynchronously, so new data may not appear in queries straight away.