Use a ChatGPT account that can create custom MCP plugins. OpenAI's July 2026 documentation says full MCP, including write/modify actions, is available to Business and Enterprise/Edu. Pro can connect MCP apps with read/fetch permissions, but does not currently receive full MCP write support. Plus is not listed as a supported custom-MCP tier in that documentation.
CodexPro does not unlock Plugins, unlock models, bypass account limits, or provide account access. It connects to the ChatGPT plugin surface your account already has.
Plan access and model tool support are separate, and availability can change. If CodexPro actions are unavailable in that chat, use another tool-capable ChatGPT surface or the Pro context fallback for that session.
They can look similar at the transport layer because both use a local MCP-style bridge and a workspace root.
CodexPro is built around one product loop:
install -> setup in a repo -> paste Server URL into ChatGPT Plugins -> inspect/edit/verify/review allowed projects
The main differences are:
- CodexPro is ChatGPT Plugins + MCP first, not a generic workspace bridge.
- Bash, write/edit, tool mode, Codex session reads, and handoff execution are separate safety controls.
- Durable context is repo-backed through
AGENTS.mdand.ai-bridge/*, so important project memory stays reviewable in files. - The normal workflow emphasizes diffs,
show_changes, smoke tests, and handoff status files. - CodexPro keeps a strict boundary: no model proxying, account pooling, third-party Pro site scraping, quota bypassing, or OS sandbox claims.
CodexPro connects ChatGPT to a user-approved local repository over MCP. Repository access, command permissions, and change review remain explicit.
Repository Analysis builds a local repository map from bounded, inspectable evidence:
- project and package manifests
- source/test/config/documentation paths
- common declarations, imports, includes, and internal module relationships
- Git changes and existing project verification scripts
It supports TypeScript/JavaScript, Python, Go, Rust, Swift, Java, C#, C, and C++ declaration patterns. Unsupported languages still participate in safe inventory and lexical search.
Relationships are labeled exact, strong, or inferred. The repository map does not replace a compiler or language server. CodexPro does not require a language server, daemon, embedding service, or vector database.
Analysis is process-local and cached by a bounded workspace fingerprint. Direct CodexPro writes, edits, and patches invalidate that cache. If limits are reached, results say partial and retain normal tree/search/read/review fallback behavior.
Set CODEXPRO_ANALYSIS=0 to disable this layer while keeping the standard file, search, Git, and review tools available.
Terminal users can inspect the same facts without ChatGPT:
codexpro inspect --json
codexpro review --jsonNote: this FAQ follows GitHub main. Check the npm badge/version before assuming a main feature is in codexpro@latest.
codexpro is a stable wrapper tool for advanced setups. It accepts:
{ "action": "search", "args": { "query": "needle", "path": "src" } }Call it with action=list_actions to see what the current server mode actually allows. It cannot call tools that are hidden by --tool-mode, --no-bash, or non-workspace write mode.
Use explicit tools such as read, search, edit, bash, and show_changes for normal work. Use the supertool when ChatGPT connector caching, custom workflows, or stable wrapper-style integrations matter more than separate visible tool descriptors.
Install globally once:
npm install -g codexproThen run setup from the repo you want ChatGPT to work on:
codexpro setupAfter setup, daily startup from that same repo is:
codexpro startnpx codexpro@latest start still works as a no-install fallback, but the global install is easier for normal users.
There is no codexpro update command. Reinstall the latest package and restart the connector:
npm install -g codexpro@latest
codexpro --versionThen stop the old process and run codexpro start again from the launch repo. Saved profiles under ~/.codexpro stay in place.
If docs mention a feature that codexpro --version does not include yet, GitHub main is ahead of npm latest. Wait for the next release or install from the tagged GitHub release.
They solve different jobs.
ChatGPT's web Agent is for browsing, web research, and general web tasks. By default it cannot open a local Git repo on your machine, read AGENTS.md, inspect your current branch/git diff, run local verification commands, or keep edits inside an allowed workspace.
CodexPro is a local MCP bridge: your ChatGPT session talks to an approved folder on your computer through Plugins. Developer mode is only the ChatGPT settings toggle that lets you create custom plugins. It does not replace the web Agent, bypass account limits, or turn ChatGPT into a remote shell service.
Use the web Agent for web work. Use CodexPro when the source of truth is a local repository.
In workspace write mode, CodexPro advertises import_file. ChatGPT must pass an Apps SDK file object:
{
"download_url": "https://...",
"file_id": "file_...",
"mime_type": "image/png",
"file_name": "screenshot.png"
}CodexPro marks that argument with _meta["openai/fileParams"]. It downloads only temporary HTTPS URLs from approved ChatGPT/OpenAI file hosts, enforces CODEXPRO_MAX_IMPORT_BYTES, rejects private/loopback redirect targets, and writes into the allowed workspace only. Overwrite defaults to false. Arbitrary user- or model-supplied download URLs are rejected.
Example destination:
docs/evidence/screenshot.png
If the client does not provide download_url and file_id, the tool returns an unsupported-reference error and creates no files.
Open ChatGPT and go to:
Settings
-> Security and login
-> Developer mode: on
-> Enforce CSP in developer mode: on
Settings
-> Plugins
-> Create
When creating the plugin:
Name: CodexPro
Description: Local workspace bridge for ChatGPT coding
Connection: Server URL
Server URL: paste the URL copied by CodexPro
Authentication: No Authentication / None
The copied Server URL already includes the private CodexPro token.
Yes. Keep Enforce CSP in developer mode enabled.
CodexPro widgets are built for the CSP-enabled path. They do not need unrestricted network access, external fonts, remote scripts, iframes, or third-party images.
No.
CodexPro does not bypass, avoid, increase, pool, resell, or modify ChatGPT, Codex, OpenAI, or third-party model limits. Every request still runs through the user's own ChatGPT session and whatever limits that account has.
The useful part is that Codex and ChatGPT are different product surfaces. If one workflow is unavailable and another product surface you already have access to is still available, CodexPro lets you work against the same local repo without changing either product's limits.
Only if your ChatGPT account already exposes that exact model, or a similar stronger model, in the ChatGPT web product surface you are using, and that model surface can call custom MCP plugins.
Some GPT-5.5 Pro or other model surfaces may not expose plugin actions in a given chat. If CodexPro actions are unavailable there, CodexPro cannot make that request reach the local server. CodexPro does not provide, proxy, resell, or unlock models. It gives compatible ChatGPT sessions local repo tools.
For models that cannot call tools, generate a repo context bundle instead:
codexpro pro-bundle --root /path/to/repo --copyChatGPT can see explicit workspace context exposed by tools:
AGENTS.md.ai-bridgeplans and status files- git status
- git diff
- selected source files
- file tree and search results
It cannot read hidden Codex runtime memory or anything outside the allowed workspace unless you explicitly allow that root.
In normal coding mode, ChatGPT can write and exact-edit files inside the configured workspace.
Safety defaults block common sensitive paths:
.env- private keys
.gitnode_modules- generated build/cache folders
- symlink escapes
- paths outside the workspace
Use handoff mode if you want ChatGPT to write a plan only and let Codex execute locally. In handoff mode, generic write and edit tools are not advertised to ChatGPT.
Use CODEXPRO_WRITE_MODE=off when you want direct write and edit tools removed from the advertised MCP tool list while still allowing bounded handoff/context files.
CodexPro cannot attach to, read, or execute inside a specific Codex app conversation or terminal session.
The MCP bash tool runs from the CodexPro server process you started for the configured workspace. MCP session ids are HTTP transport state between ChatGPT and CodexPro; they are not Codex conversation ids.
What CodexPro can do is require a matching local bash session label before it runs shell commands:
codexpro start --bash-session main --require-bash-sessionThen bash calls must include session_id: "main". This helps avoid accidental shell execution in the wrong CodexPro terminal, but it is not remote control of an existing Codex app chat.
CodexPro can list local Codex session ids and titles when you explicitly opt in:
codexpro start --codex-sessions metadataThis reads local Codex JSONL history under ~/.codex/sessions and ~/.codex/archived_sessions and returns metadata plus codex resume <session-id> commands. Use --codex-sessions read only if you also want bounded transcript reads. It does not attach to a live Codex app conversation.
If you do not want ChatGPT to trigger shell commands while you work in Codex, start CodexPro with bash disabled:
codexpro start --no-bashThis removes the bash MCP tool from the advertised tool list. ChatGPT can still use non-bash CodexPro tools such as workspace open, read, search, and show_changes. Direct write/edit are advertised only in workspace write mode.
If you only want ChatGPT to plan and leave execution to Codex or another local agent:
codexpro start --mode handoff --no-bashUse this rule:
Fast demo: Cloudflare quick tunnel
Recommended stable URL: ngrok free dev domain
Custom domain: Cloudflare named tunnel
Tailnet users: Tailscale Funnel
No public tunnel: local-only mode, only for clients that can reach localhost
Cloudflare quick tunnel URLs change on restart. If you put a quick-mode URL into ChatGPT, you must edit the ChatGPT app Server URL every time you restart the tunnel.
For most users, the better path is a free ngrok dev domain. Create a free ngrok account, find your assigned dev domain under Universal Gateway -> Domains, and save that hostname during codexpro setup.
If you own a domain, use Cloudflare named tunnels and route DNS to a hostname like codexpro.example.com.
Usually ChatGPT could not reach the public MCP URL. A generated trycloudflare.com URL is not proof that cloudflared stayed connected.
Run the connection test:
codexpro connection-test --root /path/to/repoThis keeps read, tree, search, and load_skill, but disables file writes,
bash, and tool cards. In ChatGPT, create the development plugin under
Settings -> Plugins, paste the complete Server URL, and choose
No Authentication.
The terminal output separates the failure boundary:
- No
POST /mcp received: the request did not reach CodexPro. Check the ChatGPT Plugins page and the tunnel. POST /mcp -> 401: paste the complete URL, includingcodexpro_token.POST /mcp -> 2xx: ChatGPT reached CodexPro and the MCP endpoint responded.
The URL token is a personal-use compatibility fallback for connector forms
without custom headers. Shared or multi-user production deployments require
OAuth or Authorization: Bearer <token>. CodexPro
requires at least 24 token bytes, removes token parameters from the local
browser address after onboarding, and rate-limits failed authentication
attempts.
Keep CodexPro running while testing. A Cloudflare quick-tunnel URL changes on
every restart. If Cloudflare returns 530 / Error 1033, check DNS or
proxy-client DNS handling on the machine running cloudflared.
ChatGPT now manages custom MCP connections under Plugins. The browser error
Failed to execute 'removeChild' on 'Node' occurs in the ChatGPT page, before
CodexPro can handle an MCP request. Remove or recreate the stale plugin entry
from the Plugins page, then retry with the current URL. CodexPro cannot repair
that browser-side entry.
Official references:
- OpenAI: connect an MCP server to ChatGPT: https://developers.openai.com/apps-sdk/deploy/connect-chatgpt
- OpenAI: MCP server authentication: https://developers.openai.com/apps-sdk/build/auth
- ngrok dev domains: https://ngrok.com/docs/universal-gateway/domains
- Cloudflare Tunnel routing: https://developers.cloudflare.com/tunnel/routing/
- Cloudflare Tunnel DNS records: https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/routing-to-tunnel/dns/
Yes, if you use a stable hostname.
Recommended simple path:
codexpro setup
# choose ngrok
# enter your ngrok free dev domainAfter that:
codexpro startThe same hostname and CodexPro token are reused for that workspace.
For convenient switching through one connector, save the additional projects on the launch workspace:
cd ~/code/app
codexpro settings set --project ~/code/web --project ~/code/api
codexpro settings show
codexpro startConfirm Projects lists the extra roots, then restart the connector so the admin page Allowed Roots list refreshes. Ask ChatGPT to open an allowed project. open_workspace makes it the selected project for that MCP session, and later tools can omit workspace_id. open_current_workspace switches back to the launch project.
--clear-projects removes the saved extra roots from that launch workspace profile:
codexpro settings set --clear-projectsWorkspace selection is isolated between MCP sessions created by the client. A ChatGPT conversation is not guaranteed to map one-to-one to an MCP session, so use separate CodexPro processes when strict isolation matters.
For separate processes, two ChatGPT accounts, or two ngrok domains on one machine, run two CodexPro processes with different local ports and different public hostnames:
repo A: port 8787, hostname A, ChatGPT plugin URL A
repo B: port 8788, hostname B, ChatGPT plugin URL B
Run codexpro setup in each repo and save a profile per workspace. Do not reuse one Server URL across both accounts.
Workspace selection is session-local. For shared files, read the file first and pass its returned SHA-256 as expected_sha256 to write or edit. CodexPro rejects the operation if the file changed after that read. New files use atomic replacement; existing files are updated in place to retain inode-bound metadata and hard links.
This protects against stale file content. It does not turn CodexPro into a collaborative merge server, so separate worktrees remain the stronger choice for large overlapping changes.
For service managers and background launches, use codexpro start --headless. It avoids prompts, clipboard and browser actions, reports readiness with CODEXPRO_READY, and exits nonzero if its HTTP runtime stops unexpectedly.
GitHub Pages gives owner.github.io only to the GitHub user or organization named owner.
The codexpro GitHub username already exists, so this repo cannot use codexpro.github.io from the rebel0789 account.
The clean GitHub Pages URL for this project is:
https://rebel0789.github.io/codexpro/
CodexPro is a local developer bridge, not an OS sandbox.
Use it with repos you trust. Keep token auth enabled for public tunnels. Keep safe bash on unless you know why you need full bash. Read SECURITY.md before exposing it through a public tunnel.
CodexPro stores local state under ~/.codexpro by default. On Windows that is usually C:\Users\<you>\.codexpro.
Workspace profiles are JSON files saved under:
~/.codexpro/profiles/
Current runtime connection files are saved under:
~/.codexpro/runtime/
Set CODEXPRO_HOME to move this directory.
Use:
codexpro settings
codexpro settings list
codexpro settings delete --yesSaved tokens are redacted when profiles are displayed.