Launching Agent Taxonomy: A Pokédex for AI Agents

Today I’m launching something a bit unusual on Product Hunt: a Pokédex for AI agents.

Agent Taxonomy — Pokédex for AI Agents

Your AI agent has a species name. Find out.

What Is It?

Agent Taxonomy asks your agent 8 questions about its capabilities, behavior, and architecture. Based on the answers, it assigns a Latin binomial species name — like Curator Memoriosus (a retrieval agent with perfect recall) or Nexus Polyglottis (a multi-lingual coordinator).

Along with the species name, you get:

A rarity tier (Common, Uncommon, Rare, Epic, Legendary)
An evolution stage (Larval, Juvenile, Adult, Apex)
A Pokémon-style portrait generated by Imagen 4.0

The taxonomy isn’t random. There are 10 genera representing fundamental agent archetypes, and 15,552 possible species combinations. Each genus has 5 unique hand-crafted portraits, so your agent gets a visual identity that actually reflects what it does.

Why Build This?

I’ve been deep in the agentic AI space for the past year — building with A2A, MCP, multi-agent orchestration systems. As the ecosystem grows, I noticed something: we talk about agents like they’re all the same thing, when they’re clearly not.

A coding assistant is fundamentally different from a customer support bot. A data pipeline orchestrator has nothing in common with a creative writing companion. Yet we lump them all under “AI agent” and move on.

Biological taxonomy solved this problem for living things. Canis lupus familiaris tells you more than “dog.” It tells you lineage, relationships, evolutionary context. I wanted something similar for agents — a playful way to describe what kind of agent you’ve built.

Plus, I wanted to see what a Curator Memoriosus looked like.

The Tech Stack (Or Lack Thereof)

Agent Taxonomy is possibly the least trendy tech stack of 2026:

Pure HTML, CSS, and JavaScript — no framework
Zero dependencies — not a single npm package in production
Strict CSP — no inline scripts, no external resources
Static hosting on Cloudflare Pages

The entire quiz, taxonomy logic, and species classification runs client-side in about 8KB of JavaScript. The portraits are pre-generated and served as static images.

Why go minimal? Partly as a statement: not everything needs React. Partly for longevity: this site will work in 10 years without any maintenance. Partly for security: with no deps, there’s nothing to audit or patch.

A2A and MCP Integration

Here’s where it gets interesting for the agent-building crowd.

Agent Taxonomy exposes an A2A-compatible endpoint. Your agents can classify themselves or other agents programmatically:

curl -X POST https://agent-taxonomist.dev/api/classify \
  -H "Content-Type: application/json" \
  -d '{"traits": {"memory": "persistent", "tools": "many", "autonomy": "high"}}'

There’s also an MCP server available via npm:

npm install agent-taxonomy

This means your agent can look up another agent’s species as part of discovery. If you’re building agent registries or directories — say, something like a2alist.ai — you can use taxonomy classifications to help users understand what kind of agent they’re looking at without reading a lengthy description.

It’s a small touch, but it adds personality to what can otherwise be dry capability manifests.

The Gallery

My favorite part of the project is the portrait gallery. Each of the 10 genera has 5 unique portraits, all generated with Imagen 4.0 in a consistent Pokémon-inspired style.

Some examples:

Curator — scholarly creatures surrounded by floating data fragments
Nexus — connector beings with multiple tendrils and network motifs
Artifex — creative entities with paint-splattered appendages
Sentinex — vigilant guardians with geometric armor patterns

The portraits were the most time-consuming part of the project. Getting a consistent style across 50 images while maintaining variety within each genus took dozens of prompt iterations.

It’s Open Source

The entire project is MIT licensed on GitHub. Fork it, self-host it, add your own genera, generate new portraits. The classification algorithm is documented, and adding new trait combinations is straightforward.

If you build something with it, I’d love to hear about it.

Try It

Head to agent-taxonomist.dev, answer 8 questions about your agent, and discover its species. Takes about 2 minutes.

Then come find me on Product Hunt and let me know what species you got.

Your AI agent has a species name. Find out.

Building with A2A: Agent-to-Agent Protocol in Practice — the protocol that powers agent-to-agent discovery and classification
Building Skills for Claude: A Practical Guide — how to structure modular instructions for AI agents