Examples of MCP Servers Used by Claude Plugins: TRES Finance, TestDino, and QuickDesign

Claude plugins communicate with external services through MCP (Model Context Protocol) endpoints, with concrete production implementations including TRES Finance for blockchain accounting, TestDino for automated testing, and QuickDesign for video generation.

The anthropics/claude-plugins-community repository demonstrates how MCP servers act as secure bridges between Claude's skill system and diverse backends. These servers enable standardized tool invocations without hardcoding secrets or endpoint URLs directly in skill definitions.

What Are MCP Servers in Claude Plugins?

MCP servers provide a protocol layer that abstracts external API communication. Each plugin declares its MCP server configuration—including URLs, authentication, and permissions—in a .mcp.json file, allowing skills to invoke tools by symbolic identifier rather than managing raw network requests.

This architecture separates business logic from infrastructure concerns. Skills reference server names like user-tres-finance or testdino-mcp, while the underlying MCP layer handles routing to concrete endpoints such as https://ai.tres.finance/mcp or https://mcp.testdino.com.

Real-World Examples of MCP Servers

The community repository contains three documented MCP server implementations that demonstrate different integration patterns for GraphQL and HTTP-based services.

TRES Finance MCP Server

The TRES Finance plugin bundles an MCP server at https://ai.tres.finance/mcp that exposes GraphQL tools for blockchain accounting. According to tres-finance-plugin/README.md (line 126), this server provides methods including build_query, execute, and introspect for ledger management and transaction analysis.

Skills reference this server using the identifier user-tres-finance as defined in tres-finance-plugin/.mcp.json. This configuration enables secure GraphQL operations against blockchain data without embedding API credentials in skill YAML.

TestDino MCP Server

TestDino connects to https://mcp.testdino.com to provide Playwright-based test automation capabilities. As documented in testdino/README.md (line 14), this remote MCP server exposes tools for test-run inspection, audit trails, and manual test management.

The server is configured in testdino/.mcp.json with the identifier testdino-mcp. This setup allows Claude to retrieve run reports and analyze test results through standardized tool invocations rather than direct browser automation code.

QuickDesign MCP Server

QuickDesign's MCP server operates at https://app.quickdesign.io/api/mcp, offering HTTP-based video generation and asset creation. The connector documentation in quickdesign/skills/quickdesign/references/connecting-claude-ai-via-mcp.md (line 11) describes commands like quickdesign_generate_video for automated content production.

Configured in quickdesign/.mcp.json under the identifier quickdesign, this server handles media processing workflows including video upscaling and style transfer operations through a unified interface.

How Skills Invoke MCP Servers

Skills communicate with MCP servers through YAML-defined tool invocations that specify the server identifier in the args block. This pattern appears consistently across the repository's skill definitions.

Querying Blockchain Data via TRES Finance

The following snippet from tres-wallets-upload/SKILL.md demonstrates calling the execute tool against the TRES Finance MCP server:

description: |
  Upload and onboard multiple on-chain wallets or exchange accounts into TRES.
  All GraphQL calls use the **TRES Finance MCP** server.

steps:
  - name: List available wallets
    tool: execute
    args:
      server: "user-tres-finance"
      query: |
        query getWallets { wallets { id name } }

The user-tres-finance value resolves to https://ai.tres.finance/mcp through the .mcp.json configuration, enabling secure GraphQL execution without URL hardcoding.

Retrieving Test Reports via TestDino

Skills fetch Playwright execution data by targeting the TestDino MCP server identifier:

description: Retrieve a Playwright run report from the TestDino MCP.

steps:
  - name: Get run details
    tool: get_run_report
    args:
      server: "testdino-mcp"
      run_id: "{{user_input.run_id}}"

The testdino-mcp server value routes this request to https://mcp.testdino.com, where the MCP protocol handles the HTTP communication with TestDino's backend testing infrastructure.

Generating Media via QuickDesign

For content creation workflows, skills invoke the QuickDesign MCP server using specialized media generation tools:

description: Generate a marketing video via the QuickDesign MCP.

steps:
  - name: Create video
    tool: quickdesign_generate_video
    args:
      server: "quickdesign"
      prompt: "A sleek AI-powered fitness app"
      style: "modern"

This configuration points to https://app.quickdesign.io/api/mcp, allowing Claude to trigger video rendering pipelines through the standardized MCP interface.

MCP Server Configuration Architecture

Each plugin defines its MCP server connections in a .mcp.json file at the plugin root. These configuration files map symbolic server names to concrete URLs and permission scopes, creating a portable and secure architecture.

Key configuration files include:

• tres-finance-plugin/.mcp.json – Declares the GraphQL endpoint and required OAuth scopes for blockchain operations
• testdino/.mcp.json – Configures the remote MCP endpoint for test automation tools
• quickdesign/.mcp.json – Maps the quickdesign identifier to the video generation API

Plugin manifests such as testdino/.claude-plugin/plugin.json and quickdesign/.claude-plugin/plugin.json reference these MCP definitions, ensuring skills remain agnostic of specific API implementations while maintaining secure credential management.

Summary

• MCP servers provide standardized endpoints for Claude plugins to interact with external GraphQL and HTTP APIs without embedding credentials in skill code.
• The anthropics/claude-plugins-community repository demonstrates three production patterns: TRES Finance (https://ai.tres.finance/mcp) for blockchain accounting, TestDino (https://mcp.testdino.com) for Playwright testing, and QuickDesign (https://app.quickdesign.io/api/mcp) for video generation.
• Skills invoke MCP servers by referencing symbolic identifiers (e.g., user-tres-finance, testdino-mcp) in YAML tool definitions, with actual URLs and authentication managed in .mcp.json configuration files.
• This architecture separates business logic from infrastructure concerns, enabling secure, portable plugin development across diverse backend services.

Frequently Asked Questions

What does MCP stand for in Claude plugins?

MCP stands for Model Context Protocol, sometimes referred to as Claude Plugin Server in documentation. It defines a standardized way for Claude skills to communicate with external services through configured endpoints, handling authentication and request routing automatically without exposing secrets in skill definitions.

How do I configure a custom MCP server for my Claude plugin?

Create a .mcp.json file in your plugin's root directory that maps a server identifier to a URL and required permissions. For example, the TRES Finance plugin defines its server in tres-finance-plugin/.mcp.json, allowing skills to reference user-tres-finance instead of hardcoding https://ai.tres.finance/mcp directly in YAML files.

Can Claude plugins use multiple MCP servers simultaneously?

Yes, a single plugin can declare multiple MCP servers in its .mcp.json configuration, and individual skills can target different servers within the same workflow. Each tool invocation specifies its target server in the args block using the configured identifier, enabling complex workflows that combine multiple external services.

What types of APIs can MCP servers wrap?

MCP servers can wrap any HTTP-based API, including GraphQL endpoints (like TRES Finance's blockchain interface), REST APIs, and specialized media processing services (like QuickDesign's video generation). The protocol handles the translation between Claude's tool invocations and the underlying API format, making it compatible with diverse backend architectures.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}