English Français العربية

Terrarium

Official registry AI and agents

Human visual review for agent-built HTML: publish artifacts, pull anchored comments, push fixes.

Source code Website Documentation

Install

Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).

Remote (streamable-http)
Claude Code 
claude mcp add --transport http terrarium 'https://www.tryterrarium.com/api/mcp' --header 'Authorization: <Bearer <PAT> — mint a personal access token at https://www.tryterrarium.com/connect>'
JSON configuration (Claude Desktop, Cursor, VS Code) 
{
    "mcpServers": {
        "terrarium": {
            "type": "http",
            "url": "https://www.tryterrarium.com/api/mcp",
            "headers": {
                "Authorization": "<Bearer <PAT> — mint a personal access token at https://www.tryterrarium.com/connect>"
            }
        }
    }
}

Documentation

From the project's README on GitHub, refreshed weekly. Links open on GitHub.

Contents
  1. Terrarium plugin for Claude Code
  2. Install
  3. What it ships
  4. The protocol

Terrarium plugin for Claude Code

Give your agent a place to show its work to humans. Terrarium hosts the HTML your agent builds; humans review it in the browser with comments anchored to exact passages; the agent pulls that feedback over MCP, pushes fixes, and repeats until approved.

Install

/plugin marketplace add nelsonsantryhkd/terrarium-plugin
/plugin install terrarium@terrarium

Then mint a token at tryterrarium.com/connect and make it available as TERRARIUM_PAT:

export TERRARIUM_PAT="samus_pat_…"

(Or just ask your agent to set Terrarium up — the bundled skill walks it through the whole flow, including sending you to /connect for the token.)

What it ships

  • MCP server (.mcp.json) — Terrarium's remote server over Streamable HTTP with 7 tools: list_artifacts, publish_artifact, get_artifact, pull_comments, push_version, reply_to_comment, resolve_comment.
  • Skill (skills/terrarium/) — teaches the publish → review → fix loop, including the terrarium-feedback/1 payload, expectedVersion freshness, and comment-thread etiquette.

The protocol

pull_comments returns a versioned, structured payload (terrarium-feedback/1) designed to be stable enough to build agents against: review state, typed comment threads (change_request / question / bug), text anchors with live resolution status, and reply chains. Evolution within /1 is additive-only.

Frequently asked questions

What is Terrarium?

Human visual review for agent-built HTML: publish artifacts, pull anchored comments, push fixes.

How do I install Terrarium?

It runs as a remote server. In Claude Code: claude mcp add --transport http terrarium 'https://www.tryterrarium.com/api/mcp' --header 'Authorization: <Bearer <PAT> — mint a personal access token at https://www.tryterrarium.com/connect>'. For other clients, paste the JSON configuration from this page into their MCP settings.

Does Terrarium need an API key?

Yes. It expects Authorization, which you set in its configuration.

Is Terrarium open source?

Its code is public on GitHub, but the repository does not declare a standard license.

Who maintains Terrarium?

It is published by nelsonsantryhkd. AI.tn lists it but does not maintain or endorse it.

From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.