OpenAI Codex release notes: 0.143.0 gets plugins and remote flows ready for teams

OpenAI Codex has a new stable CLI release. GitHub release rust-v0.143.0 was published on July 8, 2026, and the npm registry now shows @openai/codex with latest: 0.143.0. The alpha track has already moved to 0.144.0-alpha.4, but the practical signal is that 0.143.0 is the stable line to check first.

Codex CLI is OpenAI's local coding agent: it runs in the terminal and can read, change, and test code inside the workspace and permissions you give it. This release is not about a shiny new button. It is about something teams actually need: plugins that are easier to discover, network traffic that follows company proxy rules, better remote pairing, and clearer MCP flows.

Source: GitHub release 0.143.0, npm @openai/codex registry

OpenAI Codex release notes: what changes in 0.143.0?

0.143.0 gathers several Codex changes that were previously spread across the alpha track. For Hammer readers, the useful reading is not "upgrade everything now." It is: which parts of Codex are starting to look like operable infrastructure?

The main points:

• Remote plugins are now enabled by default, with richer catalog rows, npm marketplace sources, and visible remote/local versions.
• Codex can route authentication and Responses API traffic through macOS and Windows system proxies, including PAC and WPAD.
• codex remote-control pair exists for generating manual pairing codes from an already running daemon.
• MCP tools use tool search by default when the model and provider support it, and ChatGPT-hosted MCP servers can explicitly use session authentication.
• App-server clients get more building blocks: environment info, descendant thread listing, and history fork through a specific turn.

MCP, Model Context Protocol, is a way to connect an AI agent to external tools and data sources with clearer boundaries. A system proxy is the organization's route for network traffic, for example PAC/WPAD or a managed proxy. Both sound dry. They decide whether Codex works in a real Nordic office, school, or developer environment.

Source: GitHub release 0.143.0, PR #30297 remote plugins default, PR #31335 Responses API via system proxy, PR #29486 MCP tool search

Why it matters for Nordic teams

This is an integration release. If Codex is going to be more than a lonely terminal tab, the team needs simple answers: which plugins exist, where they come from, which version is running, how traffic leaves the machine, and who approves tool calls?

Remote plugins being on by default makes Codex more useful, but also more dependent on a working plugin routine. OpenAI's plugin documentation says the directory can be opened in the app or with /plugins in Codex CLI, that marketplace tabs switch sources, and that existing approval settings still apply after installation.

The system proxy part is just as practical. PR #31335 explicitly says the migration covers the two HTTP endpoints for /responses and /responses/compact; WebSockets, model discovery, memories, realtime, and file uploads are outside this change. That boundary belongs in your own runbook so nobody assumes all Codex traffic automatically follows the same path.

Source: OpenAI Codex plugins documentation, PR #29913 remote-control pair, PR #31335 Responses API via system proxy

Short example: use the new Codex feature

Human step first: check 0.143.0 in a test workspace and decide which plugins, MCP servers, and remote connections are in scope. Do not let the agent install packages, pair devices, or change configuration in the prompt below.

Inspect this Codex workspace for a 0.143.0 rollout. Do not edit files, install packages, pair devices, or change configuration. Return one short readiness receipt with: visible Codex version/channel, plugin sources and versions if visible, MCP servers and approval boundaries, proxy assumptions for Responses traffic, remote-control pairing runbook status, and one human decision before team rollout.

Good output should:

• separate stable 0.143.0 from alpha 0.144.0-alpha.4;
• show which plugin and MCP assumptions were actually visible;
• mention the proxy boundary without claiming all traffic is covered;
• end with a human decision, for example whether plugins should be enabled for more repo owners.

What should you test now?

I would start with one workspace where Codex is already in use. Check the version, plugin list, MCP servers, proxy constraints, and remote pairing. Save the result as a short run receipt. If something is missing, fix the routine before you give Codex more repos, more plugins, or more network access.

That is where Tool Forge fits: not as a brake, but as the way to make Codex genuinely useful. Set scoped tokens, environment variables or a secret manager, clear approval gates, redaction for sensitive output, and a run log someone on the team actually reviews.