OpenAI Codex release notes: CLI 0.139.0 brings web search to Code mode

OpenAI Codex CLI 0.139.0 is the kind of release note that looks technical at first, then changes how a coding agent should be used in real work. The main thing: Code mode can now call standalone web search directly and receive results as plaintext. For teams debugging dependencies, API docs, or MCP tools, that means less copy-paste and more reviewable research inside the workflow.

• Source: OpenAI Codex changelog, Codex CLI 0.139.0

OpenAI Codex release notes: what changed in CLI 0.139.0

OpenAI Codex CLI is OpenAI's local coding agent for the terminal. A coding agent can read code, suggest changes, and run tools in a selected project folder, but it still needs clear boundaries, tests, and human review. In 0.139.0, four practical details are worth noticing:

• Code mode can call standalone web search directly, including from nested JavaScript tool calls, and receive search results as plaintext.
• Tool and connector schemas now preserve oneOf and allOf, and large schemas keep more shallow structure when compacted. That should help richer MCP tools. MCP, Model Context Protocol, is a way to connect Codex to external tools and context such as documentation, browsers, GitHub, or Sentry.
• codex doctor now includes editor and pager environment details in the local report, while raw values are redacted in JSON output.
• codex plugin marketplace list --json now includes each marketplace source, and plugin lists can return from a cached catalog before a background refresh finishes.

• Source: GitHub release rust-v0.139.0

Human step: update and take a snapshot

If you use the npm package, the verified install line in the release notes is:

npm install -g @openai/[email protected]

Then run the checks you actually need, for example codex doctor or codex plugin marketplace list --json, and paste the relevant excerpt into Codex instead of asking the agent to guess your environment. The npm registry shows 0.139.0 as the current latest version at the time of this post.

• Source: npm registry for @openai/codex latest

Why this matters for Nordic automation teams

Web search in Code mode is useful when it leaves a trail. Ask Codex to show which search results shaped the suggestion, which files are affected, and which human needs to approve the next step. Then web research becomes part of the work receipt, not just something the agent did in the background.

The MCP change matters for the same reason. Many internal tools have rich input schemas, conditional fields, and connector rules. If Codex preserves more of that schema, it becomes easier to govern tool calls with scoped API keys, environment variables, a secret manager, approval gates, redacted output, and logs. That is the Tool Forge angle: do not just write a prompt, build a controlled run environment.

• Source: OpenAI Codex MCP documentation

Short example: use the new Codex feature

Human step first: update Codex CLI to 0.139.0, run codex doctor, and export the plugin or MCP list you want to review. Paste a short excerpt, then use this prompt:

Use Codex CLI 0.139.0. I pasted a short codex doctor report, plugin marketplace JSON, and one MCP tool schema below. Use Code mode web search only to verify current public docs for the dependency or API names you mention. Return: search sources used, schema or plugin risks, and the smallest safe next task. Do not edit files until I approve.

Good output should:

• show which web-searched sources were used, not just say research was done
• separate environment issues, plugin/MCP risks, and the code task itself
• point out whether a oneOf or allOf field may affect the tool call
• end with a small next step that requires human approval before file edits

Watch the next changelog

The latest 0.140 alpha builds are already visible on GitHub and npm, but the release text I found is only generic prerelease noise. That is not enough for a client recommendation yet. For now, 0.139.0 is the stable Codex CLI signal to test: web search in Code mode, better MCP compatibility, and more readable diagnostics.

• Source: OpenAI Codex GitHub releases