gemini-cli-mcp-slim
Thin MCP server wrapping the Gemini CLI for AI assistants, supporting workspace-aware multi-repository analysis and forward-compatible flags.
README
gemini-cli-mcp-slim
A thin, auditable MCP server wrapping the Gemini CLI.
Why
Existing Gemini MCP servers tend to either hide the underlying CLI behind opaque parameters or hard-code the supported flags so new gemini CLI options require code changes. This project takes a different approach:
- Single file, ~290 lines — auditable in one sitting
- One third-party dependency (
mcp) — minimal supply-chain surface - Forward-compatible — any new or uncommon gemini CLI flag is reachable via
extra_argswithout touching this server - Configurable binary path —
$GEMINI_CMDlets you swap or wrap the gemini binary - Workspace-aware — first-class
include_directoriesso you can analyze multiple repositories from one invocation - Transparent — every invocation logs the exact argv to stderr
Installation
Requires the gemini CLI to be installed and on $PATH (or pointed to via $GEMINI_CMD).
# Run directly without installing
uvx gemini-cli-mcp-slim
# Install from PyPI
pip install gemini-cli-mcp-slim
# Run from GitHub HEAD
uvx --from git+https://github.com/tksfjt1024/gemini-cli-mcp-slim gemini-cli-mcp-slim
Usage as an MCP server
Claude Code
claude mcp add gemini-cli-mcp-slim uvx gemini-cli-mcp-slim
Or manually in ~/.claude.json:
{
"mcpServers": {
"gemini-cli-mcp-slim": {
"type": "stdio",
"command": "uvx",
"args": ["gemini-cli-mcp-slim"]
}
}
}
Other MCP clients
Any MCP-compatible client can launch the server via stdio:
uvx gemini-cli-mcp-slim
Tools
consult_gemini
Run a single Gemini CLI invocation with full forward-compatibility.
| Parameter | Type | Description |
|---|---|---|
query (required) |
string | Prompt sent verbatim to the gemini CLI |
directory (required) |
string | Working directory (gemini cwd, defines the default workspace root) |
model |
string | Model alias (flash, pro) or full model id |
approval_mode |
enum | default / auto_edit / yolo / plan |
include_directories |
string[] | Extra workspace directories (mapped to gemini --include-directories) |
yolo |
bool | Pass --yolo (auto-approve all) |
sandbox |
bool | Pass --sandbox |
extra_args |
string[] | Raw CLI flags appended verbatim. Use to access new/uncommon gemini flags |
env |
object | Extra environment variables for the gemini subprocess |
timeout_seconds |
int | Subprocess timeout in seconds (default 600) |
consult_gemini_with_files
Same parameters as consult_gemini plus a files: string[] parameter. File paths are appended to the prompt as @path tokens (gemini's file reference syntax). Absolute paths inside directory are converted to relative @-paths.
web_search
Convenience wrapper that prepends a "use your built-in web search" instruction to the query.
Configuration
| Environment variable | Default | Purpose |
|---|---|---|
GEMINI_CMD |
gemini |
Path to the gemini CLI binary |
GEMINI_CLI_MCP_SLIM_TIMEOUT |
600 |
Default subprocess timeout in seconds |
GEMINI_CLI_MCP_SLIM_LOG_LEVEL |
INFO |
Logging level for stderr diagnostics |
Cross-repository analysis example
The killer feature: analyze multiple repositories from a single MCP invocation.
{
"name": "consult_gemini",
"arguments": {
"query": "Compare the API surface between these services",
"directory": "/path/to/service-a",
"include_directories": [
"/path/to/service-b",
"/path/to/service-c"
],
"model": "flash",
"approval_mode": "plan"
}
}
This launches gemini with --include-directories=/path/to/service-b,/path/to/service-c, giving the model read access to all three workspaces in one session.
Forward-compatibility example
If a future gemini CLI release adds a new flag (say --super-mode), you can use it immediately without updating this server:
{
"name": "consult_gemini",
"arguments": {
"query": "...",
"directory": "...",
"extra_args": ["--super-mode", "--some-other-new-flag"]
}
}
Comparison with alternatives
| Project | Lines | Deps | include_directories |
extra_args passthrough |
|---|---|---|---|---|
| gemini-cli-mcp-slim (this) | ~290 | 1 (mcp) |
yes | yes |
| eLyiN/gemini-bridge | ~400 | 1 (mcp) |
no | no |
| jamubc/gemini-mcp-tool | (TS) | many | no | partial |
Development
# Install dev dependencies
pip install -e ".[test,dev]"
# Lint
ruff check .
# Test
pytest
License
MIT © tksfjt1024
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。