Taxonomy skill for Claude Code: audit and govern your Amplitude event names

Amplitude's skill that makes Claude audit your event taxonomy, score problems and propose fixes, writing only after you confirm.

Installs: 48.7K; on skills.sh
Stars: 46; on GitHub
Updated: Oct 2026; last repo commit

taxonomy is Amplitude's rulebook for event tracking plans. It makes Claude read your taxonomy, check names against your own settings first and Amplitude's defaults second, score each issue by impact and effort, and say what to fix first. Other Amplitude skills, such as discover-event-surfaces and instrument-events, read it for naming standards.

It works through the Amplitude MCP server. It reads freely and asks you to confirm before and after states for any write. It bundles no scripts. For deletion audits it points you to the tracking-plan-audit skill.

Good for

Data and analytics owners who want a scored audit of event and property names
Engineers adding events who need Amplitude's naming and property standards
Admins planning cleanup who want a phased deprecation, not a one-step delete

Install taxonomy

Claude Code

Copy this prompt.

Install the taxonomy skill in Claude Code. Run this command:

claude plugin install amplitude

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

The plugin route installs every skill in the Amplitude plugin and the Amplitude MCP server from its .mcp.json (https://mcp.amplitude.com/mcp). In Claude Code the README tells you to authenticate when prompted. The skills CLI route copies the skill folders. For the Amplitude MCP server, use the plugin route or add https://mcp.amplitude.com/mcp to your client yourself.

From the project README. Source.

What's in the skill

When it loads

Source of truth for event taxonomy generation, data auditing, and governance best practices in Amplitude. Use when an agent needs to create, validate, audit, score, or recommend improvements to event tracking plans, naming conventions, property standards, data quality, or deprecation workflows, including checking a plan against the customer's own standard (Data → Settings naming conventions and AI Context). Covers naming rules, property standards, scoring frameworks, safe metadata operations, deprecation procedures, and AI readiness guidance.

Sections in SKILL.md

When to Use
Core Philosophy
Data Quality Lifecycle
Event Volume vs. Taxonomy Type Counts
Event States and Metadata Permissions
Custom Events, Labeled Events, and Merged Events
Protected Data Categories
Interpreting Usage Signals
AI Readiness
When Reading and Analyzing (Always Safe)
Whose Standard Applies
Auditing Against the Customer's Standard
When Writing or Updating Metadata
When Recommending Cleanup or Deprecation
When Recommending New Instrumentation
Event Naming Standards

Key facts

Author: Amplitude, Inc. (vendor's own repo; LICENSE names Amplitude, Inc. as copyright holder, README calls it the official Amplitude plugin)
License: MIT (LICENSE and README)
Installs on skills.sh: 48,682 as of Nov 2, 2026
GitHub stars: 46 for the amplitude/mcp-marketplace repo, Nov 2, 2026
What is inside: 1 file: SKILL.md (the git tree lists nothing else in the folder)
Part of: The amplitude plugin (plugins/amplitude), version 1.7.0 in the marketplace manifest
Last commit to the repo: Oct 31, 2026
Needs: An Amplitude account with API access and an MCP-compatible client (README Requirements)
Works in: Claude Code, Cursor, Codex, Gemini CLI and Claude (README)

What it includes

Principles, quality loop, event states and AI readiness
Six principles including evidence-first and confirm before writing. A four-stage loop: detect, clarify, resolve, prevent. It also covers event volume versus type counts, event states and AI readiness.
Read · Layer 1: foundational concepts
Reads Data > Settings and AI Context before judging names
Order: Data > Settings naming rules, then AI Context (project over org), then a convention the taxonomy already follows, then Amplitude's standards. If Settings and AI Context disagree, it judges by Settings and reports the conflict.
Network · Whose standard applies
Lists each violation with the rules it breaks and a fix
Each finding names the rules broken and their source and gives a compliant value. Fixes go in one change set for a tracking-plan branch, never main.
Try: “Audit my event names against our naming settings.”
Read · Audit against your standard
Edits descriptions, categories, tags and display names
Needs before and after confirmation for every write, and only confirmed items are updated. It does not write to bracket-prefixed system events, or to Unexpected or Deleted events.
Write · Write or update metadata
Recommends phased deprecation
Never presents delete, hide or block as a one-step fix. Deletion-focused audits go to the tracking-plan-audit skill.
Read · Cleanup and deprecation advice
Event, property, type, user-ID and category rules
Events use Object plus past-tense verb in Title Case. Properties use snake_case. Includes ID, flag and timestamp types, and six categories.
Read · New instrumentation standards
Scores issues on impact, event importance and effort
Three dimensions drive the order of the recommendations.
Read · Scoring and prioritizing
Metadata-only changes are allowed; block, delete and merge are not
Allowed with approval: display names, descriptions, categories, tags, official status, hiding events, Automated Tasks, AI Context, and tracking-plan edits on a branch you review.
Write · Authority boundaries
Thresholds, audit metrics and good-versus-bad examples
Detection signals with thresholds, key audit metrics, and examples of good and bad metadata.
Read · Layer 3: detection reference
Reads the connected MCP catalog at run time
Tells Claude to pick capabilities from the tool descriptions and schemas, paginate through every cursor, read the exact object before updating it, and require confirmation before any write.
Network · Amplitude MCP capability selection

From the official docs

Permissions

Medium risk

This skill reads your whole taxonomy and can edit event metadata, each write after you confirm. Approve small groups of changes at a time.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It can edit event metadata in Amplitude
It can change descriptions, categories, tags, display names, official status and hide events, and can set AI Context or edit a tracking-plan branch. SKILL.md asks for before and after confirmation, so check each change. github.com
The plugin route adds an Amplitude MCP server
The amplitude plugin ships a .mcp.json that points Claude at https://mcp.amplitude.com/mcp. You sign in to Amplitude when prompted, and the skill can then read from that project. github.com
The plugin route installs every Amplitude skill
The repo has 36 skill folders in the plugin, while the README lists 27. Use the skills CLI route in the install box if you want only the ones this page names. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	SKILL.md names no read-only setting.
Talks to	mcp.amplitude.com (the Amplitude MCP server from the plugin .mcp.json)
Sign-in and keys	Signs in to Amplitude through the MCP server when you are prompted; the README does not say where the sign-in is stored.
Scopes	Reads your event and property taxonomy, settings and AI Context through the MCP server Can update event and property metadata and tracking-plan branches after you confirm

Questions

What does the taxonomy skill do?

It gives Claude Amplitude's rules for event tracking plans. Claude can audit names and metadata, score problems, recommend cleanup in phases and apply confirmed metadata fixes. Other Amplitude skills read it as the naming source of truth.

Can it delete or block events?

SKILL.md lists blocking, deleting, merging and transforming events as not allowed. It allows metadata edits and hiding events, with your approval. For deletion audits it points to the tracking-plan-audit skill.

Will it change my Amplitude project without asking?

SKILL.md says never to call a write tool without explicit confirmation and to show before and after states first. Tracking-plan fixes go to a branch the customer reviews and merges, never to main.

How do I install it?

Use the install box above. The plugin route installs the whole Amplitude plugin and the Amplitude MCP server. The README lists Claude as supported but gives no install steps for it. The Claude tab here is a generated upload route that also needs the Amplitude MCP server added in Claude.

Is it official?

Yes, it is in the vendor's own repo. The amplitude/mcp-marketplace README calls the plugin the official Amplitude plugin for AI coding tools, and the LICENSE names Amplitude, Inc. as the copyright holder.

Does it need an Amplitude account?

Yes for audits. The README lists an Amplitude account with API access as a requirement, and SKILL.md says every finding must be grounded in data retrieved by tools.

Is the taxonomy skill safe?

It bundles no scripts and writes only metadata after you confirm. Your event names and descriptions are read into the chat, so use a project you may share with your AI provider. Our advice: ask for an audit first and approve fixes one group at a time.

Information

updated Nov 2, 2026

Repository: amplitude/mcp-marketplace
Folder: plugins/amplitude/skills/taxonomy

amplitude/mcp-marketplace skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 5

amplitude/mcp-marketplace README · github.com
marketplace.json · github.com
taxonomy SKILL.md · github.com
skills.sh listing · skills.sh
LICENSE · github.com