ci-triage-mcp
MCP server for GitHub Actions CI failure triage. It wraps the gh CLI to fetch PR checks, failed jobs, and tail/grep failed job logs with ANSI/timestamp stripping and root-cause extraction.
README
ci-triage-mcp
MCP server for GitHub Actions CI failure triage. Wraps the gh CLI to fetch PR checks, failed jobs, and tail/grep failed job logs — with ANSI/timestamp stripping and root-cause extraction. Zero runtime dependencies; stdio JSON-RPC.
Built to turn a 5–10 minute "open the run, find the failed job, scroll the raw log" ritual into a single tool call.
Tools
| Tool | Input | Returns |
|---|---|---|
get_pr_checks |
repo, prNumber |
Aggregated check rollup: failed / pending / success + merge state |
get_failed_jobs |
repo, runId |
Jobs with conclusion=failure: name, databaseId, url, timestamps |
get_job_logs_tail |
repo, jobId, lines? |
Last N log lines, ANSI + timestamp prefixes stripped |
grep_job_logs |
repo, jobId, pattern, maxMatches? |
Matching lines + 2 lines context. Use when failure is mid-log (E2E, build errors) |
summarize_pr_failure |
repo, prNumber |
End-to-end: failed checks → failed jobs → log tail → extracted error lines + file:line |
All repo inputs validated as owner/repo (alphanumerics, ., -, _). lines capped at 500, maxMatches at 200.
Install
One-click install
Click a badge → opens the install page → click the button → your IDE prompts to install.
GitHub strips custom protocol schemes (
cursor://,vscode://) from markdown links for security, so the badges route through a hosted install page on GitHub Pages with real<a href>buttons. The page also covers Claude Code, Claude Desktop, Cline, Windsurf, Zed, and LM Studio.
The install page registers the server under the name ci-triage via npx -y github:aamar-shahzad/ci-triage-mcp (works now, no npm registry publish needed). After install, restart your IDE if the tools don't appear immediately.
Manual install — per client
The server config is the same everywhere; only the file location differs.
Cursor — ~/.cursor/mcp.json:
{
"mcpServers": {
"ci-triage": {
"command": "npx",
"args": ["-y", "github:aamar-shahzad/ci-triage-mcp"],
"env": {}
}
}
}
VS Code / GitHub Copilot — ~/.vscode/mcp.json (or workspace .vscode/mcp.json):
{
"mcpServers": {
"ci-triage": {
"command": "npx",
"args": ["-y", "github:aamar-shahzad/ci-triage-mcp"],
"env": {}
}
}
}
Claude Desktop — claude_desktop_config.json:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"ci-triage": {
"command": "npx",
"args": ["-y", "github:aamar-shahzad/ci-triage-mcp"],
"env": {}
}
}
}
Claude Code (CLI):
claude mcp add ci-triage npx -- -y github:aamar-shahzad/ci-triage-mcp
Cline / Roo Code / Continue / Windsurf — same mcpServers shape as Cursor, in their respective MCP config files:
- Cline:
~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json(macOS) - Windsurf:
~/.codeium/windsurf/mcp_config.json - Continue:
~/.continue/config.jsonunderexperimental.mcpServers
Zed — ~/.config/zed/settings.json:
{
"context_servers": {
"ci-triage": {
"command": "npx",
"args": ["-y", "github:aamar-shahzad/ci-triage-mcp"],
"env": {}
}
}
}
LM Studio — deep link:
lmstudio://add_mcp?name=ci-triage&config=eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsImdpdGh1YjphYW1hci1zaGFoemFkL2NpLXRyaWFnZS1tY3AiXSwiZW52Ijp7fX0=
Option C — clone + run directly
For local development or to avoid npx cold-install delay:
git clone https://github.com/aamar-shahzad/ci-triage-mcp.git ~/src/ci-triage-mcp
Then in your MCP config:
{
"mcpServers": {
"ci-triage": {
"command": "node",
"args": ["/Users/you/src/ci-triage-mcp/src/index.mjs"]
}
}
}
Regenerate install links
node scripts/gen-install-links.mjs # prints markdown
node scripts/gen-install-links.mjs --json # prints raw links as JSON
Authentication
No token stored in MCP config. Uses your existing gh CLI authentication (keyring / config file from gh auth login). The server shells out to gh; nothing else.
Requirements:
ghCLI on PATH — install from https://cli.github.com- Authenticated:
gh auth login(scopes:repo,workflowfor CI logs) - Node ≥ 20
On initialize, the server probes gh --version and emits a _warning in serverInfo if gh is missing.
Why
Generic GitHub MCPs return raw JSON and stop at the check rollup. This server is opinionated about the triage workflow:
- Failed-check discovery —
get_pr_checksfilters toconclusion=FAILUREonly. - Job drilldown —
get_failed_jobsmaps a run to its failed job IDs (the part you usually copy by hand). - Clean logs —
get_job_logs_tailstrips ANSI escapes and the2026-07-09T20:34:56.5281699Ztimestamp prefix that buries the actual line. - Mid-log grep —
grep_job_logsfinds errors anywhere in the log with context. E2E / build failures live mid-log, not at the tail. Tail-only tools miss them. - Root-cause extraction —
summarize_pr_failurechains all of the above and extracts error lines +file:linein one call.
Example: a 725-line Cypress job log → grep_job_logs with pattern AssertionError|Timed out returns 1 match with 2 lines context, surfacing Expected to find element: [data-testid="inviteUserButton"], but never found it in one call.
Smoke test
node scripts/smoke.mjs
Runs tools/list + a grep_job_logs call against a known public failed job. Requires gh auth.
Development
git clone https://github.com/aamar-shahzad/ci-triage-mcp.git
cd ci-triage-mcp
node src/index.mjs # starts stdio JSON-RPC server
No build step. Pure ESM .mjs, no dependencies.
License
MIT
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。