# What Are the Agent Infrastructure Plugins for Claude? MCP Server Integrations Explained

> Discover agent infrastructure plugins for Claude. Learn how these MCP server integrations connect Claude to external services like image generation and test management via HTTP endpoints.

- Repository: [Anthropic/claude-plugins-community](https://github.com/anthropics/claude-plugins-community)
- Tags: deep-dive
- Published: 2026-09-04

---

**Agent infrastructure plugins for Claude are remote MCP (Model-Control-Protocol) server integrations that enable Claude agents to invoke external services like image generation, test management, and text simplification through standardized HTTP endpoints defined in [`.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/.claude-plugin/plugin.json) configuration files.**

The `anthropics/claude-plugins-community` repository hosts specialized agent infrastructure plugins that extend Claude's capabilities beyond local execution. These plugins provide the foundational connectivity that allows Claude Code and other agent environments to securely interact with third-party services through the Model-Control-Protocol (MCP), eliminating the need for local CLI installations while maintaining strict permission controls.

## Understanding Agent Infrastructure Plugins

Agent infrastructure plugins differ from standard local tools because they expose a **remote MCP server** rather than executing code on the user's machine. In the `claude-plugins-community` repository, these plugins are located in dedicated directories containing a [`.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/.claude-plugin/plugin.json) file that declares the HTTP endpoint, authentication requirements, and available tool methods.

Each plugin follows a consistent pattern: the agent connects to the MCP endpoint once via OAuth 2.1, caches the authentication token, and subsequently routes all tool invocations through the remote server. This architecture enables Claude to interact with services like QuickDesign or TestDino from any environment—including the web UI at claude.ai—without requiring local dependencies.

## Available Agent Infrastructure Plugins

The `anthropics/claude-plugins-community` repository currently maintains three primary agent infrastructure plugins that expose production MCP endpoints.

### QuickDesign Image and Video Generation

The **QuickDesign** plugin provides a remote HTTP MCP endpoint at `app.quickdesign.io/api/mcp` that enables agents to execute generative AI workflows without local GPU resources. According to the source code in [`quickdesign/.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/quickdesign/.claude-plugin/plugin.json), this plugin exposes methods for image generation, video generation, and advertisement creation pipelines.

When a Claude agent invokes `quickdesign_image_generate`, the request travels to the QuickDesign MCP server, which processes the generation task and returns asset URLs that the agent can embed in responses. The plugin handles OAuth 2.1 authentication during initial setup, after which all tool calls are automatically signed and routed through the secure endpoint.

### TestDino Test Management

The **TestDino** plugin, configured in [`testdino/.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/testdino/.claude-plugin/plugin.json), exposes a read-only MCP server for accessing test run histories, flaky test detection, and manual testing assets. This infrastructure plugin enables natural language queries like "show me the last flaky test for login" by translating them into `testdino_list_runs` and `testdino_fetch_run` tool invocations.

The MCP server returns structured JSON payloads containing test logs, screenshots, and metadata that the agent can analyze and summarize for the user. Because the plugin operates as a remote server, development teams can query test results from any Claude interface without installing the TestDino CLI locally.

### Eli5 Text Simplification

The **Eli5** plugin supplies a lightweight "explain in plain language" service through its MCP endpoint defined in [`eli5/.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/eli5/.claude-plugin/plugin.json). When agents encounter technical jargon or complex documentation, they can invoke `eli5_explain` to send passages to the remote server and receive concise, jargon-free rewrites suitable for non-technical stakeholders.

This infrastructure plugin is particularly valuable for cross-functional teams where Claude acts as a translator between technical and business domains, processing text through the `eli5_explain` tool and integrating the simplified output directly into conversation threads.

## How to Invoke Agent Infrastructure Plugins

Claude agents interact with these infrastructure plugins using asynchronous tool calls that route through the MCP protocol. Once a user has installed a plugin via `claude plugin marketplace add`, the agent can invoke methods using the following patterns:

```python

# QuickDesign – generate a photorealistic image

await claude.run_tool(
    "quickdesign_image_generate",
    prompt="A futuristic city skyline at sunset",
    style="photorealistic",
    width=1024,
    height=768,
)

```

```python

# TestDino – retrieve recent test execution history

await claude.run_tool(
    "testdino_list_runs",
    limit=5,
)

```

```python

# Eli5 – simplify technical architecture descriptions

await claude.run_tool(
    "eli5_explain",
    text="""
    The service follows a CQRS pattern where commands are dispatched
    to a message bus that persists events to an event store.
    """,
)

```

Each invocation returns a JSON payload specific to the service—image URLs from QuickDesign, test metadata from TestDino, or simplified text from Eli5—that the agent parses and presents to the user.

## Configuration and Source Files

The infrastructure plugins rely on declarative configuration files that define the MCP contract. Reviewing these files reveals the exact capabilities and permission scopes available to agents:

- **[`quickdesign/.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/quickdesign/.claude-plugin/plugin.json)** – Declares the MCP base URL, required OAuth scopes, and the complete tool set including `quickdesign_image_generate` and video generation methods.
- **[`testdino/.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/testdino/.claude-plugin/plugin.json)** – Specifies the TestDino MCP endpoint and enumerates read-only tools for accessing test runs and flaky test histories.
- **[`eli5/.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/eli5/.claude-plugin/plugin.json)** – Defines the Eli5 MCP endpoint and the single `eli5_explain` tool with its expected text parameter schema.

These JSON files reside in hidden `.claude-plugin` directories within each plugin folder, serving as the source of truth for the Claude plugin marketplace when registering available agent capabilities.

## Summary

Agent infrastructure plugins for Claude provide secure, remote MCP server integrations that expand agent capabilities without requiring local software installations:

- **QuickDesign** enables generative media creation through `quickdesign_image_generate` and related tools.
- **TestDino** offers test infrastructure access via `testdino_list_runs` for quality assurance workflows.
- **Eli5** provides text simplification through the `eli5_explain` tool for cross-functional communication.
- Each plugin is defined in a [`.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/.claude-plugin/plugin.json) file that specifies the MCP endpoint and authentication requirements.
- These plugins support OAuth 2.1 flows and execute in remote environments, making them accessible from Claude Code, Claude Cowork, and the web interface.

## Frequently Asked Questions

### What is an MCP server in Claude plugins?

An **MCP (Model-Control-Protocol) server** is a remote HTTP endpoint that exposes standardized tool methods for Claude agents to invoke. According to the `anthropics/claude-plugins-community` source code, MCP servers defined in files like [`quickdesign/.claude-plugin/plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/quickdesign/.claude-plugin/plugin.json) handle authentication, execute the requested operations on remote infrastructure, and return structured JSON responses that agents can process and display to users.

### How do I install agent infrastructure plugins?

Install these plugins using the Claude plugin marketplace commands as documented in the repository's [`README.md`](https://github.com/anthropics/claude-plugins-community/blob/main/README.md). After adding the marketplace source, run `claude plugin marketplace add <plugin-name>` to register the MCP endpoint. The initial connection triggers an OAuth 2.1 flow to authorize the agent's access to the remote service.

### What is the difference between agent infrastructure plugins and local tools?

**Agent infrastructure plugins** route tool calls to remote MCP servers via HTTP, enabling access to services like QuickDesign without local installations. Local tools, by contrast, execute code directly on the user's machine through subprocess calls or Python functions. Infrastructure plugins are essential when services require specialized hardware (GPUs), proprietary APIs, or centralized data stores that cannot be replicated locally.

### Are agent infrastructure plugins secure for production use?

Yes, these plugins implement **OAuth 2.1 authentication** and permission scoping as defined in their respective [`plugin.json`](https://github.com/anthropics/claude-plugins-community/blob/main/plugin.json) files. Each plugin requests only the necessary permissions for its function—such as read-only access for TestDino—and all communications occur over HTTPS to the declared MCP endpoints. Users maintain control through explicit connection approvals, and agents cannot access the infrastructure without valid, user-granted tokens.