GitHub Flow MCP
A policy-aware MCP server for GitHub and GitHub Actions that enables safe AI-assisted infrastructure workflows—inspecting repositories, preparing branches and pull requests, and constrained remote mutations behind explicit preview-bound approval tokens.
README
GitHub Flow MCP
A policy-aware Model Context Protocol server for GitHub and GitHub Actions workflows. It is designed for AI-assisted infrastructure work where agents can inspect repositories and prepare changes, while remote mutations remain constrained by explicit preview-bound approval.
Why this exists
AI agents are useful at collecting context and preparing repetitive Git operations, but infrastructure workflows still need reviewability and guardrails. This MCP exposes a deliberately small GitHub surface:
- repository, branch, pull request, review, and Actions inspection
- local Git inspection without silent branch/commit/push/merge operations
- branch and pull-request previews
- short-lived, single-use confirmation tokens bound to the exact preview
- optional Basecamp linkage
- optional repository-specific label and reviewer policy
- no merge tool, no branch deletion, no branch-protection changes, and no API-created commits
Safety model
Remote write tools are not registered unless:
GITHUB_WRITE_ENABLED=true
When enabled, branch creation, pull-request creation, and PR comments still require a confirmation_id returned by their matching preview. The token is short-lived, single-use, and bound to the exact payload. If a branch source SHA, PR body, label set, reviewer set, or comment changes, a new preview is required.
[!NOTE] The MCP enforces preview-before-write and exact payload binding. The MCP host should only forward the
confirmation_idafter actual human approval.
Installation
uv tool install git+https://github.com/thiagorchaves/github-flow-mcp.git
For development:
git clone https://github.com/thiagorchaves/github-flow-mcp.git
cd github-flow-mcp
uv sync --extra dev
Configuration
cp .env.example .env
chmod 600 .env
Use a fine-grained GitHub token with the minimum repositories and permissions needed. Typical permissions are Metadata read, Contents read/write for branch creation, Pull requests read/write, Issues read/write for labels/comments, Actions read, and Members read only when reviewer availability checks are required.
Organization-specific policy belongs in your local .env, not in the public repository. For example, you can require Basecamp links, configure a production branch marker, or require an AI-review label for selected repositories.
Kiro / MCP host example
{
"mcpServers": {
"github-flow": {
"command": "uvx",
"args": ["--from", "git+https://github.com/thiagorchaves/github-flow-mcp.git", "github-flow-mcp"],
"env": {
"GITHUB_TOKEN": "${GITHUB_TOKEN}",
"GITHUB_DEFAULT_OWNER": "${GITHUB_DEFAULT_OWNER}",
"GITHUB_INITIALS": "tc",
"GITHUB_DEFAULT_BASE_BRANCH": "main",
"GITHUB_LOCAL_REPOS_ROOT": "${HOME}/Projects",
"GITHUB_WRITE_ENABLED": "false",
"GITHUB_CONFIRMATION_TTL_SECONDS": "300"
},
"autoApprove": []
}
}
}
Tools
Read and diagnostics
healthcheckget_authenticated_userget_repositorylist_branchesget_collaborator_permissioninspect_local_repositorylist_pull_requestsget_pull_requestlist_pull_request_fileslist_pull_request_reviewslist_workflowslist_workflow_runs
Preview
preview_branch_namepreview_local_commitpreview_remote_branchpreview_pull_requestpreview_pull_request_comment
Remote writes, registered only when enabled
create_remote_branchcreate_pull_requestadd_pull_request_comment
Basecamp handoff
build_basecamp_pr_update
This tool only generates Markdown. A Basecamp MCP can independently preview and publish that update.
Recommended workflow
- Retrieve the work-item context.
- Inspect the local repository without modifying it.
- Preview the branch and show it to the user.
- After approval, use the preview token to create the remote branch if needed.
- Edit and commit locally using your normal signed Git workflow.
- Push the branch.
- Preview the pull request.
- After approval, use its token to open the PR.
- Let humans review the PR and apply the infrastructure change through the normal delivery process.
Development
uv sync --extra dev
uv run ruff format --check .
uv run ruff check .
uv run mypy github_flow_mcp
uv run pytest
Security
Never commit .env, GitHub tokens, internal repository data, company-specific paths, or private work-item URLs. See SECURITY.md.
License
MIT. See LICENSE.
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。