MCP and Plugins Integration - Connecting Local and Remote Tools Safely

Model Context Protocol (MCP) connects Codex to external tools and context. A Plugin packages Skills, Apps, and MCP server configuration as an installable workflow.

MCP compared with Plugins

Capability	MCP	Plugin
Main purpose	Provide tools and context	Distribute a complete workflow
Contents	Servers, tools, resources	Skills, Apps, MCP, presentation metadata
Example	Search official docs or operate Figma	A complete docs or security workflow

MCP Connection Types

Codex supports local STDIO servers and remote Streamable HTTP servers. HTTP connections can use bearer tokens or OAuth.

Configuration lives in ~/.codex/config.toml or a trusted project .codex/config.toml. The codex mcp command can also manage servers.

Before connecting, review the provider, transmitted data, read and write permissions, credential storage, tool side effects, and network need.

Plugin Composition

Plugins appear in the Codex app Plugin Directory and the CLI /plugins browser. Existing approval settings still apply after installation, while data sent to an external service is governed by that service’s terms and privacy policy.

Safe Operation

Start with read-only access, enable only required tools, keep credentials out of Git, require approval for external writes, and disable unused Plugins.

Next, Memories design separates local recall from permanent policy.

See the references for the external specifications and background sources used on this page.[1][2][3]

References

1. OpenAI, Codex documentation
2. OpenAI, OpenAI API documentation
3. Model Context Protocol, What is the Model Context Protocol?