MCP Servers

What is MCP?

The Model Context Protocol (MCP) acts as a universal adapter for AI tools. It allows Codex to interact securely with remote APIs, local databases, and third-party services without exposing your credentials or data.

Think of MCP servers as plugins that extend what Codex can do beyond reading and writing files.

How MCP Works

• MCP Server — A lightweight process that exposes tools and data to the AI agent
• MCP Client — The Codex CLI, which connects to configured servers
• Tools — Specific actions the server exposes (e.g., "read pull request", "query database")
• Resources — Data sources the server can read from (e.g., database schemas, API docs)

Configuring MCP Servers

Create or edit your mcp.json configuration file. This can be project-level or global:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
      }
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "DATABASE_URL": "postgresql://user:pass@localhost:5432/mydb"
      }
    }
  }
}

GitHub MCP Server

The GitHub integration lets Codex interact with your repositories:

# Install the GitHub MCP server
codex mcp install github-server

Once configured, Codex can:

• Read Pull Requests — Analyze PR diffs and provide review comments
• Comment on Issues — Add analysis or suggestions to GitHub Issues
• Browse repository files — Read code from other repos for reference
• Check CI status — See if builds and tests are passing

Database MCP Server

Connect a read-only database tool so Codex can write accurate SQL:

# Install PostgreSQL MCP server
codex mcp install postgres-server

# Install MySQL MCP server
codex mcp install mysql-server

# Install Redis MCP server
codex mcp install redis-server

With a database MCP server, the agent queries your live schema — it knows exactly which tables exist, what columns they have, and their data types. No more guessing column names or writing invalid queries.

⚠️ Important: Always use read-only database credentials for MCP servers. This prevents the agent from accidentally modifying production data.

Listing and Managing MCP Servers

# List all configured MCP servers
codex mcp list

# Test a server connection
codex mcp test github

# Remove a server
codex mcp remove redis-server

What's Next

In the next episode, we'll learn about delegating tasks to the cloud — how Codex can act as a project manager, breaking large objectives into sub-tasks and dispatching them to multiple cloud agents.