CodeRabbit CLI MCP
Enables AI code reviews through the CodeRabbit CLI by exposing a run_review tool that executes CodeRabbit with full flag coverage for analyzing committed, uncommitted, or all changes in a repository.
README
coderabbit-cli-mcp
English documentation lives here. A full Japanese version is available in README.ja.md.
run_review is the only MCP tool this server exposes. When the CodeRabbit CLI is missing or unauthenticated, the same tool response explains how to bootstrap it and reminds Codex to report back to the user instead of acting autonomously.
Quickstart (Codex CLI)
Published package (npx)
codex mcp add coderabbit-cli-mcp \
-- "npx" \
"-y" \
"coderabbit-cli-mcp"
Development checkout (local tsx)
codex mcp add coderabbit-cli-mcp \
--env TSX_BANNER=false \
-- "$(pwd)/node_modules/.bin/tsx" \
"$(pwd)/src/server.ts"
After registration you can inspect tools with codex mcp call coderabbit-cli-mcp tools/list or trigger run_review via codex mcp call.
Codex config (client-side)
~/.codex/config.toml controls how Codex launches MCP servers. Start with the minimal entry above, then add optional knobs:
[mcp_servers.coderabbit-cli-mcp]
command = "npx"
args = ["coderabbit-cli-mcp@latest"]
tool_timeout_sec = 1800 # allow up to 30 minutes (default 60)
# startup_timeout_sec = 60
[mcp_servers.coderabbit-cli-mcp.env]
# CODERRABBIT_MCP_LOCK_MODE = "plain"
# CODERRABBIT_MCP_LOCK_TYPE = "uncommitted"
# CODERRABBIT_MCP_LOCK_CONFIG_FILES = "[\".coderabbit.yaml\"]"
# CODERRABBIT_TOOL_TIMEOUT_SEC = "1800"
- Uncomment the optional lines you need. Keep
command/argsas-is for a published install, or swap them for the localtsxcommand from the quickstart section. tool_timeout_secshould be ≥600 whenever reviews may run longer than a minute.CODERRABBIT_MCP_LOCK_MODE/_TYPEtake plain strings.CODERRABBIT_MCP_LOCK_CONFIG_FILESmust be a JSON array (CodeRabbit’s-cflag accepts multiple files, so list all required config paths there).CODERRABBIT_TOOL_TIMEOUT_SECshould match the value you assign totool_timeout_secso this server can confirm the configuration.- These environment variables are not accepted via MCP tool arguments; Codex’s environment is the only place to set them.
- Because the environment carries these values, no extra server-side config files are required.
Provided Tool
| Tool | Role | Key arguments |
|---|---|---|
run_review |
Executes coderabbit with full flag coverage. If the CLI is missing or unauthenticated, the same response includes setup guidance and instructs Codex to ask the user. |
mode, type, base, baseCommit, cwd, extraArgs[] |
run_review arguments
| Argument | Type | Description | Default |
|---|---|---|---|
mode |
interactive|plain|prompt-only |
Mirrors the CodeRabbit CLI output format. | plain |
type |
all|committed|uncommitted |
Selects the diff scope passed to CodeRabbit. | all |
base |
string |
Overrides the base branch (--base). |
unset |
baseCommit |
string |
Overrides the base commit (--base-commit). |
unset |
cwd |
string |
Directory from which the CLI runs; must exist. | server process.cwd() |
configFiles |
string[] |
Pre-configured via CODERRABBIT_MCP_LOCK_CONFIG_FILES (JSON array). |
unset |
extraArgs |
string[] |
Appended verbatim to the CLI argv (use sparingly). | unset |
--no-coloris always appended so MCP clients receive plain output.- Only
mode/type/configFilescan be pre-configured through Codex; all other arguments must be supplied per-tool invocation.
Security & Operating Notes
- The server never attempts to install CodeRabbit or run
coderabbit auth login. Follow the emitted guidance and execute commands manually in your own shell. - Codex (or any MCP client) must not run additional commands on its own when an error occurs. Report the guidance back to the user and wait for explicit instructions before retrying.
- Windows native is unsupported; WSL2 instructions are provided instead. PATH modifications also remain manual.
- The MCP server does not read Codex’s
AGENTS.mddynamically. Keep governance rules inAGENTS.md, and rely on the environment variables above to enforcemode/type.
Development
npm install
npm run build
npm run dev
CLI Smoke Tests
# List tools through the MCP inspector
npx @modelcontextprotocol/inspector \
--cli node dist/server.js \
--method tools/list
# Call run_review (plain output against uncommitted changes)
npx @modelcontextprotocol/inspector \
--cli node dist/server.js \
--method tools/call \
--tool-name run_review \
--tool-arg mode=plain \
--tool-arg type=uncommitted
# List report:// resources for recent runs
npx @modelcontextprotocol/inspector \
--cli node dist/server.js \
--method resources/list
# List prompts
npx @modelcontextprotocol/inspector \
--cli node dist/server.js \
--method prompts/list
npm Publish Checklist
npm run build
npm pkg fix
npm version patch # or minor / major as needed
npm publish --access public
npm view coderabbit-cli-mcp version
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。