MCP Catalog

MCP Catalog Skill

Purpose

Comprehensive reference for Model Context Protocol (MCP) servers — official, community, and Agent Studio integrated. Use this skill when selecting, configuring, or building MCP servers for Claude Code projects.

When to Invoke

Skill({ skill: 'mcp-catalog' });

Invoke when:

• Selecting an MCP server for a specific capability
• Configuring MCP servers in Claude Code settings
• Building a new MCP server integration
• Auditing existing MCP server usage

Official Anthropic / MCP Reference Servers

Community MCP Servers (Curated)

Developer Tools

Data & Analytics

Cloud & Infrastructure

Communication

Claude Code Settings Configuration

stdio Transport (Most Common)

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/directory"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "${GITHUB_TOKEN}"
      }
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "POSTGRES_URL": "${DATABASE_URL}"
      }
    }
  }
}

SSE Transport (HTTP-based)

{
  "mcpServers": {
    "remote-server": {
      "type": "sse",
      "url": "https://your-mcp-server.example.com/sse",
      "headers": {
        "Authorization": "Bearer ${API_TOKEN}"
      }
    }
  }
}

Environment Variable Injection Pattern

Never hardcode credentials in settings.json. Use environment variable references:

{
  "env": {
    "API_KEY": "${MY_SERVICE_API_KEY}"
  }
}

Variables are resolved from the shell environment where Claude Code runs.

Agent Studio MCP Integration

Agent Studio skills that wrap MCP servers:

Adding a New MCP Server

Step 1: Identify the Server

# Search npm registry
npm search "@modelcontextprotocol/server-"

# Browse community list
# https://github.com/modelcontextprotocol/servers

Step 2: Configure in settings.json

Edit .claude/settings.json to add the mcpServers entry (see templates above).

Step 3: Test the Connection

# Verify Claude Code loads the MCP server
claude --mcp-debug

Step 4: Create an Agent Studio Skill (Optional)

If the MCP server is used frequently, wrap it in an Agent Studio skill:

# Use the skill-creator skill
Skill({ skill: 'skill-creator' })

The skill should:

• Document the MCP server's tools and resources
• Provide usage examples
• Define when to invoke
• List required environment variables

Security Considerations

Credential Management

• Store all secrets in .env (never in settings.json directly)
• Use ${ENV_VAR} interpolation in settings.json
• Rotate tokens regularly; use minimal-scope PATs

File System Access

• Restrict filesystem server to specific directories only
• Never grant root (/) or home (~) access
• Use read-only mode when write access is not needed

Network MCP Servers

• Validate SSL certificates on SSE connections
• Use bearer token auth for remote MCP servers
• Audit tool calls from remote servers — treat as untrusted input

Least Privilege

• Only register MCP servers that the agent actually needs
• Remove unused MCP server registrations
• Prefer local stdio over remote SSE when possible

MCP Protocol Reference

Tool vs Resource vs Prompt

Naming Convention in Claude Code

MCP tools appear as: mcp__<serverName>__<toolName>

Example: mcp__github__create_issue, mcp__filesystem__read_file

Discovery

To list all available MCP tools in a session:

// Agents can discover available MCP tools via AvailableAgents() or tool search
Skill({ skill: 'tool-search' });

Related Skills

• mcp-builder — Build a new MCP server from scratch
• github-ops — GitHub workflow using MCP
• github-mcp — Direct GitHub MCP operations
• figma — Figma design access via MCP
• webmcp-browser-tools — Browser automation MCP