Convert MCP JSON
to Codex TOML.

Purpose-built MCP conversion

A generic JSON-to-TOML converter cannot understand that an MCP server name must become a table path, or that a bearer-token environment reference can become Codex’s bearer_token_env_var. This converter understands the MCP client formats and applies those mappings deliberately.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/projects"],
      "env": { "LOG_LEVEL": "info" }
    }
  }
}

Is converted to:

[mcp_servers.filesystem]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/projects"]

[mcp_servers.filesystem.env]
LOG_LEVEL = "info"

Supported MCP fields

• Local stdio servers using command and args
• Per-server environment variables
• Remote Streamable HTTP server URLs
• Literal HTTP headers and environment-backed headers
• Bearer-token environment-variable references
• Codex startup and tool timeout values

Why the result is safer to review

The converter reports definitions that contain both a command and a URL, insecure remote HTTP endpoints, direct shell execution, legacy SSE transports and values that resemble real credentials. A passing check means the configuration shape is plausible; it does not guarantee the MCP server’s code is trustworthy.

Where to save the TOML

For a user-wide OpenAI Codex configuration, merge the generated tables into ~/.codex/config.toml. For a project-scoped configuration, use .codex/config.toml inside a trusted repository. Do not overwrite unrelated Codex settings when adding the generated MCP tables.