rebel0789/codexpro

CodexPro – Give ChatGPT safe, local access to your code repositories

What it is – CodexPro is a small Node‑JS server that acts as a custom MCP (Model‑Client‑Plugin) for OpenAI’s ChatGPT. When you run it on your machine and register it as a plugin inside ChatGPT, the model can call a set of tools that let it read, search, edit, and run checks on the files of any repository you explicitly allow. All analysis stays on your computer; nothing is uploaded to a third‑party service.

Why it matters – The recent “code‑assistant” wave (GitHub Copilot, ChatGPT Code Interpreter, etc.) often runs in the cloud, sending your source code to remote servers. CodexPro flips that model: the LLM stays in OpenAI’s cloud, but the heavy‑weight code‑intelligence (language detection, symbol indexing, impact analysis) runs locally, giving you tighter control over privacy, security, and reproducibility.

Quick start (from the README)

Prerequisites
- Node 20+ installed
- A ChatGPT account that can add custom MCP plugins
- An HTTPS endpoint that points to your machine (e.g., a Cloudflare tunnel, ngrok, or Tailscale Funnel)
Install the CLI
```
npm install -g codexpro
```
Set up a repo
```
cd /path/to/your/project
codexpro setup
```
This creates a local configuration and prints the HTTPS URL you will give to ChatGPT.
Add the plugin in ChatGPT
- Turn on Developer mode in Settings → Security & login.
- Go to Settings → Plugins → + next to Search plugins.
- Create a plugin called CodexPro and paste the URL from step 3.
- Choose No Authentication / None (the token is already embedded in the URL).
Start the server
```
codexpro start   # runs the MCP server for the current repo
```
The server now listens for tool calls from your ChatGPT session.

Core capabilities (as described in the README)

Capability	What ChatGPT can do through CodexPro
Workspace inspection	`inspect_workspace` builds a lightweight index of languages, project types, entry points, symbols, and internal relationships.
Search	`search` supports plain‑text, regex, and intent‑driven queries like `symbol`, `references`, and `impact`.
Edit	`write`, `edit`, or guarded `apply_patch` let the model propose and apply code changes inside the allowed roots.
File import	`import_file` accepts attachments from the ChatGPT Apps SDK and writes them into the repo.
Command execution	A safe subset of Bash commands can be run (`bash` tool) after you enable it.
Diff & impact analysis	`show_changes` returns the diff, highlights affected areas, dependent code, related tests, risk signals, and suggested verification commands.
Planning & handoff	The model can write a hand‑off plan under a `.ai-bridge` directory and export a context bundle for chats that cannot call tools.

Built‑in repository intelligence

Recognises declarations in TypeScript/JavaScript, Python, Go, Rust, Swift, Java, C#, C, C++. Other languages get lexical search and safe inventory.
No external embedding service, vector DB, or language‑server daemon is required – everything is cached locally and keyed to a workspace fingerprint.
Results include a coverage and inference strength metric instead of a false sense of certainty.

Safety & security defaults

Explicit workspace boundaries – you decide which directories are exposed; everything else is blocked.
Token‑protected tunnels – public HTTPS tunnels require a random 24‑byte token (stored in ~/.codexpro/http-token).
Write mode is hidden unless you start the server with --mode workspace.
Safe Bash is the default; dangerous commands are filtered.
Blocked paths include .env, private keys, .git, build caches, etc.
Attachments are only accepted from approved HTTPS hosts via the ChatGPT Apps SDK.

Typical workflow

Open a workspace – codexpro start (or codexpro start --root /path/to/repo).
In ChatGPT, ask the model to open_workspace for that repo.
The model can then:
- Search for a function definition (search symbol MyClass.myMethod).
- Propose a change (edit src/foo.py), review the diff (show_changes), and apply it (apply_patch).
- Run a test command (bash npm test).
- Write a hand‑off document summarising what was done.
When finished, stop the server or start another instance for a different repo.

Extensibility & modes

Multiple projects – a single CodexPro process can serve several repos (codexpro settings set --project …).
Tool‑mode selection – --tool-mode minimal, full, or custom card sets (CODEXPRO_TOOL_CARDS=1).
Headless / handoff / pro – different preset configurations for CI‑style runs, record‑keeping, or full interactive use.

Installation & updates

npm install -g codexpro          # install globally
codexpro --version               # verify
npm install -g codexpro@latest    # upgrade

Configuration lives under `~/.codexpro`; updating the binary does not erase your saved profiles.

Where to learn more

Website / docs – https://rebel0789.github.io/codexpro/
Roadmap, FAQ, Security guide – linked in the repo (ROADMAP.md, FAQ.md, SECURITY.md).
Source – the repository includes CI, smoke‑test, stress‑test, and release scripts for contributors.

Bottom line

CodexPro is a genuine, open‑source bridge that lets ChatGPT act as a local code‑assistant without exposing your proprietary source. It provides searchable, language‑aware indexing, safe edit/patch workflows, and optional command execution, all gated behind explicit repository permissions and HTTPS‑tunnel authentication. Ideal for developers who want the convenience of LLM‑driven coding while retaining full control over their code and environment.