# Spark History MCP — Investigate Jobs via Tools > Kubeflow’s Spark History Server MCP + `shs` CLI for job analysis, failures, and comparisons; verified 168★, pushed 2026-05-13. ## Install Merge the JSON below into your `.mcp.json`: ## Quick Use ```bash # MCP server (README): uvx --from mcp-apache-spark-history-server spark-mcp # Or install with pip: pip install mcp-apache-spark-history-server spark-mcp # CLI quickstart (README): shs setup config > config.yaml ``` ## Intro Kubeflow’s Spark History Server MCP + `shs` CLI for job analysis, failures, and comparisons; verified 168★, pushed 2026-05-13. **Best for:** Spark teams who want repeatable investigations from an agent (MCP) or scripts (CLI) **Works with:** Spark History Server; MCP server runs on port 18888 and supports streamable-http/stdio (README) **Setup time:** 12-30 minutes ### Key facts (verified) - GitHub: 168 stars · 59 forks · pushed 2026-05-13. - License: Apache-2.0 · owner avatar + repo URL verified via GitHub API. - README-backed entrypoint: `uvx --from mcp-apache-spark-history-server spark-mcp`. ## Main - Use `shs` for quick, deterministic inspection; use MCP when you want an agent to run multi-step investigations across apps and stages. - Keep config explicit: README uses `shs setup config > config.yaml` and expects you to set your History Server URL there. - Choose transport by deployment: streamable HTTP is convenient for remote clients; stdio is simple for local setups (README). - Use comparisons to avoid guesswork: README links a real-world example of comparing two benchmark runs and highlights failure investigation commands. ### Source-backed notes - README says the project provides two interfaces: an MCP server (`spark-mcp`) and a standalone CLI (`shs`). - README shows running the MCP server directly via `uvx --from mcp-apache-spark-history-server spark-mcp` and mentions PyPI publishing. - README config shows an MCP port default of 18888 and transport options `streamable-http` or `stdio`. ### FAQ - **Do I need MCP if I only want scripts?**: No — use `shs` CLI directly; MCP is for agent-driven investigations (README positioning). - **Where do I set the Spark History Server URL?**: In `config.yaml`; README generates it via `shs setup config > config.yaml`. - **What port does the MCP server use?**: README defaults to port 18888 and supports transport configuration. ## Source & Thanks > Source: https://github.com/kubeflow/mcp-apache-spark-history-server > License: Apache-2.0 > GitHub stars: 168 · forks: 59 --- ## Quick Use ```bash # MCP server (README): uvx --from mcp-apache-spark-history-server spark-mcp # Or install with pip: pip install mcp-apache-spark-history-server spark-mcp # CLI quickstart (README): shs setup config > config.yaml ``` ## Intro Kubeflow 出品的 Spark History Server MCP + `shs` CLI：用于作业分析、失败排查与运行对比，适合把排障变成可复用工具链；已验证 168★，更新于 2026-05-13。 **Best for:** 希望用 agent（MCP）或脚本（CLI）复用 Spark 排障流程的团队 **Works with:** 需要 Spark History Server；MCP 默认 18888 端口，支持 streamable-http/stdio（README） **Setup time:** 12-30 minutes ### Key facts (verified) - GitHub：168 stars · 59 forks；最近更新 2026-05-13。 - 许可证：Apache-2.0；作者头像与仓库链接均已通过 GitHub API 复核。 - README 中可对照的入口命令：`uvx --from mcp-apache-spark-history-server spark-mcp`。 ## Main - 快速确定性检查用 `shs`；需要多步推理与聚合结论时用 MCP 让 agent 调度工具。 - 配置要显式：README 用 `shs setup config > config.yaml` 生成配置，并要求写入 History Server URL。 - 按部署选择 transport：远程/调试优先 streamable HTTP，本地接入可用 stdio（README）。 - 用对比减少拍脑袋：README 给出对比两次运行的示例，并提供失败排查的命令方向。 ### Source-backed notes - README 写明提供两种接口：MCP server（`spark-mcp`）与独立 CLI（`shs`）。 - README 给出 `uvx --from mcp-apache-spark-history-server spark-mcp` 直接运行方式，并说明发布在 PyPI。 - README 配置示例包含 MCP 默认端口 18888 与 `streamable-http/stdio` 传输选项。 ### FAQ - **只想写脚本还需要 MCP 吗？**：不需要；README 定位 `shs` 是给工程师/脚本用，MCP 适合 agent 调度。 - **Spark History Server 的 URL 写在哪？**：写在 `config.yaml`；README 通过 `shs setup config > config.yaml` 生成。 - **MCP 默认端口是多少？**：README 默认 18888，且可在配置中调整 transport/端口。 ## Source & Thanks > Source: https://github.com/kubeflow/mcp-apache-spark-history-server > License: Apache-2.0 > GitHub stars: 168 · forks: 59 --- Source: https://tokrepo.com/en/workflows/spark-history-mcp-investigate-jobs-via-tools Author: MCP Hub