commit-brief
MCP server that turns git commit history into concise standup summaries, exposing tools to summarize and list commits with filters like repo, author, and time range.
README
commit-brief
Turn yesterday's git log into a one-paragraph standup summary.
commit-brief reads your git history (commit messages, authors, branch refs,
per-file change stats — never diffs) and asks an LLM for a terse standup
digest. One command, two interfaces:
commit-brief— standup digest at your terminal.commit-brief mcp— MCP server, so your agents (Claude Code, Codex, ...) can call the same core as a tool.
Install
Requires Python 3.11+ and uv (or pipx).
# global install, any machine — CLI only:
uv tool install git+https://github.com/ahm3d-karim/commit-brief
# with the MCP server (needs the mcp dependency):
uv tool install "git+https://github.com/ahm3d-karim/commit-brief[mcp]"
# from a local checkout instead:
uv tool install ".[mcp]"
# pipx equivalent:
pipx install "git+https://github.com/ahm3d-karim/commit-brief[mcp]"
Installs one command: commit-brief. Then set your API key:
export ANTHROPIC_API_KEY=sk-ant-... # or add it to your shell profile
Model default: claude-sonnet-4-20250514 (override with COMMIT_BRIEF_MODEL).
CLI
# from inside any repo — yesterday's commits, one paragraph
commit-brief
# other repos / windows / filters
commit-brief --repo ../other-repo --since '3 days ago'
commit-brief --since '7 days ago' --author 'Alice'
commit-brief --author 'Alice' --author 'Bob' --bullets
# no API key? see exactly what the LLM would receive
commit-brief --dry-run
# raw structured commits (hash, author, subject, body, refs, per-file stats)
commit-brief --json
--since/--until accept anything git accepts: yesterday, 3 days ago,
2026-08-01. --author is repeatable (OR). Exit codes: 0 ok / no commits,
2 git or API error.
MCP server
The server exposes the same core as two tools: summarize_standup and
list_commits (params: repo, since, until, author, bullets,
dry_run — dry_run returns the prompt at zero API cost).
Run it (needs the [mcp] extra installed):
commit-brief mcp
# from a source checkout: uv run --extra mcp python -m commit_brief.mcp_server
Register with Claude Code:
claude mcp add commit-brief -- commit-brief mcp
Then just ask: "summarize yesterday's commits in this repo" or "what did each dev ship last week?" — the agent calls the tool with the right args.
Self-test the server end to end (spawns it, lists tools, exercises both
tools against a repo — pass a path or set CBR_TEST_REPO; defaults to the
current directory):
commit-brief mcp-test
commit-brief mcp-test /path/to/repo
Design notes
- Messages + stats, never diffs. The LLM sees subjects, bodies, authors,
branch refs, and per-file
(+a -d)counts. Cheap, and code never leaves the repo. - Git subprocess hygiene: git is spawned with
stdin=DEVNULLandGIT_PAGER=cat. Without this, a git spawned inside an MCP stdio server on Windows/MSYS2 gets SIGTERM'd (rc 143) because it inherits the protocol pipe / console state. This is the fix — don't remove it. mcpis pinned<2because 2.x replaced FastMCP with a new API; the stdio protocol is identical, so 1.x keeps the server code simple.
Testing
uv run --extra test pytest # core + CLI suite (unit tests, no API needed)
commit-brief mcp-test # end-to-end MCP smoke test (spawns the server)
CI runs the pytest suite on every push (GitHub Actions, .github/workflows/ci.yml).
The MCP smoke test is intentionally excluded from pytest — it spawns servers
and exits the process — so it stays a self-serve command.
Project layout
commit_brief/core.py git log parsing + prompt building + LLM call
commit_brief/cli.py one command: standup (default) + mcp / mcp-test
commit_brief/mcp_server.py FastMCP wrapper (2 tools)
commit_brief/mcp_test.py end-to-end MCP smoke test (used by mcp-test)
scripts/test_mcp_client.py thin shim over commit_brief.mcp_test
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。