ReviewLens MCP Server
Enables secure, explainable AI-assisted GitHub pull-request review by exposing seven read-only tools that collect deterministic evidence from PRs, separating retrieval from interpretation.
README
ReviewLens MCP
A secure, explainable MCP server for AI-assisted GitHub pull-request review.
ReviewLens exposes seven typed, read-only tools that collect evidence from pull requests. It separates deterministic evidence collection from model interpretation, treats repository content as untrusted data, and includes a credential-free simulated provider and web demo.

Why it exists
AI review assistants often collapse retrieval, interpretation, and action into one opaque step. ReviewLens keeps those concerns separate so tool calls are auditable, contracts are testable, and no model can mutate a repository through this server.
Product tour
The deterministic demo completes a realistic pull-request review without requiring a GitHub token. It exposes the collected evidence, related tests, risk level, and any instruction-like content found in untrusted repository data.

The tool surface is intentionally small: seven operations with explicit inputs and one responsibility each. The architecture keeps MCP at the adapter boundary and isolates review logic from GitHub access.

The security model is visible rather than implied: least privilege, bounded evidence, untrusted-content handling, and explicit failures. A restrained pixel-art signature connects the project to its creator without competing with the engineering content.

Quick start — demo mode
python -m venv .venv
# Windows: .venv\Scripts\activate
# macOS/Linux: source .venv/bin/activate
python -m pip install -e ".[dev]"
pytest
reviewlens-demo
Open http://127.0.0.1:8000. The demo always uses deterministic fixture data.
Run the MCP server
reviewlens-mcp
The default transport is stdio.
Tools
| Tool | Purpose |
|---|---|
| list_pull_requests | List open pull requests |
| get_pull_request | Fetch PR metadata |
| get_pull_request_files | Collect changed files and patches |
| get_pull_request_diff | Return a bounded, redacted diff |
| search_repository_code | Search repository code |
| find_related_tests | Locate tests connected to changed paths |
| build_review_report | Assemble deterministic structured evidence |
Live mode
Copy .env.example to .env, set REVIEWLENS_MODE=live, and optionally configure a fine-grained token or GitHub App installation token with read-only Contents and Pull requests permissions. Tokens remain server-side.
Architecture and security
See docs/architecture.md, docs/security.md, and docs/adr/. The MVP contains no write-capable provider method, MCP tool, or demo endpoint.
Verification
ruff check .
mypy src
pytest
Documentation
- Spanish: README.es.md
- Recording guide: docs/demo-recording.md
- Limitations: docs/limitations.md
Author
Built by Paula García Fernández.
Copyright
Copyright © 2026 Paula García Fernández. All rights reserved. The repository is viewable as a personal portfolio and technical demonstration; reuse requires prior written permission. See COPYRIGHT.md. Audio is intentionally absent from v1.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。