Introducing the Agent Registry: Share and Discover Anthropic Managed Agent Templates

What is the Agent Registry?

The Agent Registry is a community marketplace for Anthropic Managed Agent configurations - hosted at 21st.dev/community/agents.

An Anthropic Managed Agent is a structured config - YAML or JSON - that defines how a Claude agent behaves: which model it uses, what system prompt it follows, and which MCP servers it connects to. You build them in the Anthropic Console or via the CLI.

The problem: every team ends up writing the same DevOps triage agent, the same research assistant, the same customer support bot from scratch. The Agent Registry fixes this. Think of it as npm for Claude agent configs - publish yours in seconds, discover what others have built, and copy any agent into your own workflow with one click.

Why share agent configurations?

A well-crafted agent config represents real engineering work:

System prompts take iteration. You test edge cases, refine tone, add guardrails. A good prompt for a code review agent is 50+ lines of carefully tuned instructions.
MCP server combinations require setup. Connecting GitHub + Linear + Slack for a triage agent means figuring out auth, scopes, and which tools to expose. Someone has already done this.
Tool selection matters. Knowing which MCP tools to enable (and which to disable) for a given use case is tribal knowledge that should be shared.

Without a registry, this work stays locked inside individual teams. With it, the best configs surface, get bookmarked, and get reused across the community.

How to publish an agent

Publishing takes under a minute:

Export your config - Open your Managed Agent in the Anthropic Console and copy the YAML or JSON config.
Go to 21st.dev - Navigate to 21st.dev/community/agents and click Publish.
Paste your config - Drop in your YAML or JSON. The form instantly previews your agent: name, model, connected MCP servers, system prompt.
Let AI fill the details - Click Next and the system auto-generates a short description, suggests a category and tags, and writes a full markdown README explaining what your agent does and how to use it. Everything is editable.
Submit for review - Your agent goes through a quick moderation check. Once approved, it appears in the registry with a public profile page under your username.

Here's what a typical agent config looks like:

name: pr-review-agent
model: claude-sonnet-4-6
tools:
  - type: mcp_toolset
    server_label: github
    server_url: https://mcp.github.com
  - type: mcp_toolset
    server_label: linear
    server_url: https://mcp.linear.app
system_prompt: |
  You are a senior code reviewer. For each PR:
  1. Check for security issues, performance problems, and style violations
  2. Create Linear issues for anything that needs follow-up
  3. Post a summary comment on the PR with your findings
  Be concise. Focus on what matters.

How to discover and use agents

Browse by category

Agents are organized into 9 categories:

Coding - code review, refactoring, PR automation
Data Analysis - data pipelines, analytics, reporting
DevOps - deployment, monitoring, infrastructure
Content - writing, editing, content generation
Research - web research, knowledge synthesis
Customer Support - ticket handling, FAQ bots
Workflow - cross-tool process automation
Monitoring - alerting, observability, incident response
Multi-Agent - orchestrators that coordinate other agents

You can also filter by MCP server - find all agents that use GitHub, or Stripe, or Slack - and search by keyword.

Copy and deploy

When you find an agent you want, you have two options:

Copy prompt - One click copies a self-contained markdown document with the full system prompt, the CLI command to recreate the agent, and step-by-step setup instructions. Paste it into Claude Code or your terminal and you're running.

Copy config - Grab the raw config in your preferred format and import it directly into the Anthropic Console.

Export formats

Every agent in the registry can be exported in 5 formats:

YAML - The native Anthropic Console format:

name: support-agent
model: claude-sonnet-4-6
tools:
  - type: mcp_toolset
    server_label: slack

JSON - For programmatic use:

{
  "name": "support-agent",
  "model": "claude-sonnet-4-6",
  "tools": [{ "type": "mcp_toolset", "server_label": "slack" }]
}

CLI - One-liner to create the agent:

ant agents create --name support-agent --model claude-sonnet-4-6

Plus TypeScript and Python SDK snippets for embedding agents in your applications.

Get started

The Agent Registry is live and open to the community.

Browse agents - Discover production-ready Anthropic Managed Agent configs across 9 categories
Publish your agent - Share your config in under a minute. AI writes the docs for you.

If you've built something useful with Claude and MCP servers, the registry is the place to share it. Every published agent gets a public profile, appears in search, and can be bookmarked by the community.

Explore

Build