OpenAI’s Codex CLI added MCP server support this week. With this release, all three major-lab CLI agents (Claude Code, Codex, Jules announced) are aligning on MCP. The protocol becomes effectively universal.
For tool builders and users invested in MCP, this is the announcement that confirms the standard.
What this enables
Codex CLI users can now:
- Use the same MCP servers that work with Claude Code
- Access the growing MCP ecosystem
- Switch between Codex and Claude Code without changing integrations
- Build custom MCP servers and use them across major-lab tools
The portability is real. A Postgres MCP server doesn’t care which AI tool is calling it. The integrations developed for one work for all.
The market context
Six months ago, MCP was Anthropic-aligned. Cline used it; Claude Code shipped with it; the community was small but growing.
Today:
- Anthropic: full MCP support
- Google (Jules): announced support, beta upcoming
- OpenAI (Codex CLI): MCP support shipped
This is the trifecta. The protocol is now universal across the major labs.
What this means for users
For users on Codex CLI:
- The same MCP servers used by Claude Code users now work
- Existing setups can leverage broader ecosystem
- Tool choice becomes less locked-in to MCP support
For tool builders:
- Targeting multiple major-lab tools is easier
- One MCP server, multiple deployments
- Audience for MCP servers grows
For the ecosystem overall:
- Standard protocols win; lock-in concerns decrease
- Innovation moves to other dimensions (UX, model quality, integrations)
- Users have more freedom to switch tools
What’s not affected
A few things this doesn’t change:
Anthropic’s continued lead in MCP maturity. The Anthropic-built MCP infrastructure is more polished than the newer entries. They’ve been at it longer.
MCP server quality varies. A protocol exists; that doesn’t mean every server is well-built. Audit the servers you depend on.
OpenAI’s other tools (Cursor via OpenAI models, ChatGPT, etc.) aren’t directly affected. Those products use OpenAI’s own integration patterns.
A specific use case
I’ve been using a custom MCP server for our internal billing system. Previously it worked only with Claude Code. With this update, I can also use it from Codex CLI.
The result: I can pick the better tool for each task without losing my billing integration. For tasks where Claude is stronger, Claude Code. For tasks where GPT-4 is stronger, Codex. The infrastructure follows.
This is genuine flexibility. Pre-MCP, I’d have rebuilt the integration for each tool.
What I’d watch
A few things to track:
- MCP server quality across providers (does one provider’s implementation lag?)
- Protocol evolution as multiple providers contribute (faster iteration; possible incompatibilities)
- Whether closed-tool ecosystems try to build alternatives
- Smaller AI tools adopting MCP (becomes table stakes for any agent)
The next year should see continued MCP maturity. The current state is “standard exists; adoption is broad.” The next state is “standard works seamlessly across implementations.”
The open question
The remaining major holdout: Cursor doesn’t support MCP natively. Cursor’s chat panel uses different integration patterns.
Whether Cursor adopts MCP in 2026 will affect the protocol’s ubiquity. If yes, MCP is genuinely universal across editor and CLI tools. If no, the editor/CLI distinction creates a fragmentation point.
My guess: Cursor will eventually adopt some form of MCP support. The competitive pressure plus the ecosystem benefits make it likely.
Worth caring about?
For Codex CLI users: yes. Your tool just gained access to a meaningful ecosystem.
For users on Claude Code: indirect benefit. Your investments in MCP servers become more portable.
For other users: pay attention. The pattern of cross-vendor tool standards is one of the more healthy market dynamics.
Closing
A small announcement with broader implications. The MCP protocol is now genuinely universal across the major labs’ CLI tools. The ecosystem accelerates from here.
For tool builders and users invested in MCP, this is the milestone that justifies the bet. Long-term, MCP looks like a real standard rather than a vendor-specific protocol.
The next year should bring more MCP servers, more integrations, more reliable cross-tool experiences. The investments made now compound.