Alison Aquinas logoAlison's LLM Plugins

mcp-integration

Included in pluginshared-skillsView on GitHub ↗

Files

SKILL.mdagentsreferences

Install

Install the containing plugin
/plugin install shared-skills@llm-skills
Invoke this skill after installation
/shared-skills:mcp-integration
Download mcp-integration-skill.zip
This skill is bundled inside shared-skills. Install the plugin once, then Claude Code can use any of its included skills. Browse the full plugin repository at github.com/alisonaquinas/llm-shared-skills.

SKILL.md


name: mcp-integration description: > Client integration phase for MCP server development. Use when connecting an MCP server to an MCP-compatible agent or desktop client, configuring the mcpServers block in settings.json, adding a server to a desktop client, testing with a real client, verifying tool visibility in the client UI, troubleshooting connection failures, or configuring environment variables for server startup. Covers settings.json configuration, environment variable passing, and connection troubleshooting.

MCP Integration

Connect the server to a real client and verify tool visibility.

Intent Router

Load reference files on demand — only when the corresponding topic is active:

  • references/client-config-patterns.md — Load when configuring mcpServers in settings.json, writing Claude Desktop config, or setting environment variables
  • references/connection-troubleshooting.md — Load when the client cannot connect, tools are not visible, or the connection drops intermittently

Quick Start — Client Configuration

Add to ~/.claude/settings.json (user scope) or .claude/settings.json (project scope):

{
  "mcpServers": {
    "my-server": {
      "command": "node",
      "args": ["/absolute/path/to/dist/index.js"],
      "env": {}
    }
  }
}

For a Python server:

{
  "mcpServers": {
    "my-server": {
      "command": "python",
      "args": ["/absolute/path/to/server.py"]
    }
  }
}

Always use absolute paths. Relative paths fail when the client working directory differs from the project root.

Verification After Configuration

After saving the config:

  1. Start a new conversation in the agent
  2. Ask the agent to list available tools or check connected MCP servers
  3. Confirm the server name appears and tools are listed
  4. Call one tool to verify end-to-end connectivity

If tools are absent, load references/connection-troubleshooting.md.

Environment Variable Passing

Pass secrets and configuration via the env block in settings.json:

{
  "mcpServers": {
    "my-server": {
      "command": "node",
      "args": ["/path/to/dist/index.js"],
      "env": {
        "API_KEY": "sk-...",
        "DATABASE_URL": "postgres://..."
      }
    }
  }
}

Values in settings.json are stored in plaintext. For sensitive secrets, have the server read from a .env file at startup using dotenv or python-dotenv.

SSE Server Integration

For servers using HTTP/SSE transport, use the url field instead of command:

{
  "mcpServers": {
    "remote-server": {
      "url": "http://localhost:3001/sse"
    }
  }
}

Gate

mcp-integration is complete when:

  • The server appears in the client's connected MCP servers list
  • All declared tools are visible in the client UI
  • At least one tool call succeeds end-to-end through the real client

Proceed to mcp-validation only after tools are confirmed visible and callable in the target client.

Safety Notes

Use absolute paths in command and args — relative paths cause "spawn ENOENT" errors when the client's working directory does not match the expected location. After editing settings.json, start a fresh session for the new config to take effect.

Resource Index

Reference FileLoad When
references/client-config-patterns.mdInitial client configuration; Claude Desktop; multiple servers
references/connection-troubleshooting.mdAny connection failure; tools not visible; parse errors
← Back to marketplace