# raye-deng/open-code-review [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/raye-deng/open-code-review  
**GitHub Stars:** 40  
**npm Downloads (last month):** 61  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/raye-deng-open-code-review

## Description
AI code quality gate detecting hallucinated packages, phantom dependencies, stale APIs, and AI-specific code defects. MCP Server + CLI + CI/CD integration.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "open-code-review": {
    "command": "npx",
    "args": ["-y","@opencodereview/cli"]
  }
}
```

## Documentation

## What raye-deng/open-code-review MCP server does

The raye-deng/open-code-review MCP server exposes Open Code Review's code-quality checks to AI development environments through Model Context Protocol. Its tools are `ocr_scan`, `ocr_heal`, and `ocr_explain`: the first evaluates a codebase, the second proposes or applies automated corrections, and the third explains detected issues.

The underlying project targets defects that conventional linters may not identify, including imports for packages that do not exist, APIs that are no longer current, contradictions spanning files, unnecessary abstractions, and some security anti-patterns. Language-specific checks are documented for TypeScript and JavaScript, Python, Java, Go, and Kotlin. Package verification covers registries including npm, PyPI, Maven, and Go modules.

## How it works

A scan can use a fast structural mode or the deeper L3 workflow. The deeper workflow combines structural checks, embedding-based semantic analysis, and an LLM pass for cross-file coherence, logic defects, confidence scoring, and suggested fixes. Results include a quality score, severity information, and a pass or fail status against a configured threshold.

LLM-assisted analysis can run through Ollama or an OpenAI-compatible endpoint. This lets teams choose a local model or provide a remote model endpoint using the CLI's provider options. The MCP layer makes the review functions available to the connected client; the repository also offers a standalone CLI and CI/CD usage separately.

## Setup and configuration

Run the raye-deng/open-code-review MCP server with:

```bash
npx @opencodereview/mcp-server
```

For Claude Desktop, add an MCP entry whose command is `npx` and whose arguments request the package with `-y`. The README gives the same configuration pattern for Cursor, Windsurf, and VS Code Copilot MCP settings. The server is therefore suited to clients that can launch a standard command-based MCP server.

LLM settings are configured through the project's broader Open Code Review configuration and provider options. Ollama examples use models such as `qwen3-coder`; OpenAI-compatible examples specify an API base, model, and API key. A fast scan does not require AI, while deep analysis does require a configured model provider.

## Tools and capabilities

- `ocr_scan`: run the code-quality gate against source code.
- `ocr_heal`: generate AI-assisted fixes, with the CLI supporting dry runs before changes are applied.
- `ocr_explain`: provide more context about a reported issue.
- Detect hallucinated imports, stale APIs, security patterns, over-engineering, and selected language-specific problems.
- Use local or OpenAI-compatible LLM providers for deeper analysis.

The project also supports terminal, JSON, SARIF, and HTML output through its CLI, plus GitHub Actions and GitLab CI workflows. Those integrations are part of the wider project rather than separate MCP tools.

## Limitations and notes

The raye-deng/open-code-review MCP server is an interface to the Open Code Review project, not a general-purpose code editor or repository host. Review generated changes before applying them; the project documents a dry-run option for its CLI healing workflow.

The repository is licensed under BUSL-1.1. The README names Claude Desktop, Cursor, Windsurf, VS Code Copilot, Claude Code, and Codex as compatible or supported environments, while the MCP setup shown uses a command-based launch configuration.

_Full upstream README: https://allmcps.com/mcp/raye-deng-open-code-review/readme_