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

Server	Package	Capability	Transport
`filesystem`	`@modelcontextprotocol/server-filesystem`	Read/write local files with configurable roots	stdio
`github`	`@modelcontextprotocol/server-github`	GitHub repos, PRs, issues, search	stdio
`gitlab`	`@modelcontextprotocol/server-gitlab`	GitLab projects, MRs, pipelines	stdio
`google-drive`	`@modelcontextprotocol/server-gdrive`	Google Drive file access and search	stdio
`google-maps`	`@modelcontextprotocol/server-google-maps`	Geocoding, directions, place search	stdio
`postgres`	`@modelcontextprotocol/server-postgres`	PostgreSQL read access with schema inspection	stdio
`sqlite`	`@modelcontextprotocol/server-sqlite`	SQLite read/write with business intelligence	stdio
`slack`	`@modelcontextprotocol/server-slack`	Slack channels, messages, user management	stdio
`memory`	`@modelcontextprotocol/server-memory`	KV-based persistent memory graph	stdio
`puppeteer`	`@modelcontextprotocol/server-puppeteer`	Browser automation and web scraping	stdio
`brave-search`	`@modelcontextprotocol/server-brave-search`	Brave Search API — web and local results	stdio
`fetch`	`@modelcontextprotocol/server-fetch`	HTTP fetch with robots.txt compliance	stdio
`sequential-thinking`	`@modelcontextprotocol/server-sequential-thinking`	Dynamic chain-of-thought reasoning	stdio
`aws-kb-retrieval`	`@modelcontextprotocol/server-aws-kb-retrieval`	AWS Knowledge Base retrieval via Bedrock	stdio

Community MCP Servers (Curated)

Developer Tools

Server	Source	Capability
`exa`	`exa-labs/exa-mcp-server`	AI-powered web search (semantic)
`linear`	`linear-mcp-server`	Linear issue tracking and project management
`jira`	community	Jira issue read/write, sprint management
`notion`	community	Notion pages, databases, search
`obsidian`	community	Obsidian vault read/write
`sentry`	community	Sentry error tracking, issues, events
`datadog`	community	Datadog metrics, logs, monitors
`figma`	community	Figma design file access, component export

Data & Analytics

Server	Source	Capability
`bigquery`	community	Google BigQuery read queries
`mongodb`	community	MongoDB document CRUD and aggregation
`redis`	community	Redis key operations, pub/sub
`elasticsearch`	community	Elasticsearch search, index management
`snowflake`	community	Snowflake SQL queries

Cloud & Infrastructure

Server	Source	Capability
`aws`	community	AWS resource management (EC2, S3, Lambda)
`gcp`	community	Google Cloud resource management
`cloudflare`	`@cloudflare/mcp-server-cloudflare`	Workers, KV, R2, D1 management
`vercel`	community	Vercel deployments, projects, domains
`kubernetes`	community	Kubernetes cluster management

Communication

Server	Source	Capability
`discord`	community	Discord server management and messaging
`telegram`	community	Telegram bot messaging
`twilio`	community	SMS, voice, WhatsApp via Twilio
`sendgrid`	community	Email sending via SendGrid

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:

Skill	Underlying MCP	Agent Studio Path
`github-ops`	`@modelcontextprotocol/server-github`	`.claude/skills/github-ops/`
`github-mcp`	`@modelcontextprotocol/server-github`	`.claude/skills/github-mcp/`
`figma`	Figma MCP server	`.claude/skills/figma/`
`google-workspace`	Google Drive + Sheets MCP	`.claude/skills/google-workspace/`
`slack-notifications`	Slack MCP	`.claude/skills/slack-notifications/`
`linear-pm`	Linear MCP	`.claude/skills/linear-pm/`
`jira-pm`	Jira MCP	`.claude/skills/jira-pm/`
`arxiv-mcp`	ArXiv search	`.claude/skills/arxiv-mcp/`
`webmcp-browser-tools`	Browser tools MCP	`.claude/skills/webmcp-browser-tools/`

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

Concept	Purpose	Agent Interaction
Tool	Execute actions (write file, call API)	Agent calls `mcp__serverName__toolName()`
Resource	Read data (file contents, DB records)	Agent reads `mcp://serverName/resource/path`
Prompt	Reusable prompt templates	Agent requests prompt template by name

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

mcp-catalog

Files6

SKILL.md.claude/skills/mcp-catalog/SKILL.md