openEuler MCP Toolkit
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@openEuler MCP Toolkit check the current memory and swap usage"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
openEuler MCP Toolkit
面向 openEuler/Linux 的只读系统观测与操作系统算法实验 MCP Server。它把内存、文件系统、进程与 CPU 信息封装为类型化工具,让支持 MCP 的大模型客户端能够调用可靠、可测试、结果规模受控的系统能力。
这个仓库是 MCP Server,不内置大模型。工具选择和自然语言解释由连接它的 MCP 客户端完成。
项目特点
12 个正式工具,覆盖内存、文件系统、进程和 CPU 调度。
实时系统观测与算法仿真采用独立模块和输出类型。
Pydantic 输入输出模型,由 MCP 自动生成并校验 Schema。
文件工具受允许目录限制;不提供删除文件、结束进程等写操作。
长时间采样支持取消和进度通知。
纯算法测试、系统服务测试和真实 stdio MCP 协议测试。
20 条自然语言评测任务,不绑定模型厂商或 API Key。
Related MCP server: MCP ProcFS Server
工具列表
模块 | 工具 | 类型 | 说明 |
内存 |
| 实时观测 | RAM、Swap 和 |
内存 |
| 实时观测 | 进程 RSS、VMS 和前 N 项内存映射 |
内存 |
| 实时观测 | 有界时间窗口内的内存趋势 |
内存 |
| 算法仿真 | FIFO、LRU、CLOCK、OPT |
文件系统 |
| 实时观测 | 分区、容量和 inode |
文件系统 |
| 实时观测 | 受控目录内的文件分布 |
文件系统 |
| 实时观测 | 轮询文件大小和时间戳变化 |
文件系统 |
| 算法仿真 | 连续、链式、索引分配 |
调度 |
| 实时观测 | 有深度和节点上限的进程树 |
调度 |
| 实时观测 | 系统或单进程上下文切换增量 |
调度 |
| 算法仿真 | FCFS、SJF、RR、Priority |
调度 |
| 实时观测 | CPU 各状态时间占比 |
快速开始
需要 Python 3.11 或 3.12。推荐在 openEuler/Linux 上运行;macOS 可运行算法和多数 psutil 工具,但没有 /proc 数据。
python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e ".[dev]"启动 stdio Server:
openeuler-mcpstdio 是协议通道,直接运行后没有交互式提示属于正常现象。可以运行协议示例:
python examples/smoke_client.py使用 MCP Inspector:
mcp dev src/openeuler_mcp/server.py通用客户端配置见 examples/mcp-config.json。请把命令改为虚拟环境中 openeuler-mcp 的绝对路径,并替换允许访问的目录。
文件访问安全
文件分析默认只能访问 Server 的启动目录。通过系统路径分隔符配置多个允许目录:
export OPENEULER_MCP_ALLOWED_ROOTS="/var/log:/home/user/safe-data"工具拒绝允许目录以外的路径。
扫描时跳过符号链接。
不要授权包含私钥、浏览器配置、Cookie 或其他敏感数据的目录。
MCP 客户端可能将工具结果发送给所配置的模型服务,请同时检查客户端的数据策略。
三条演示链路
1. 系统内存检查
查看当前内存和交换区使用情况,并说明数据来源。
客户端应调用 get_memory_info,然后区分 psutil 数据与 Linux /proc/meminfo 数据。
2. 进程内存分析
分析 PID 1234 的内存占用,只列出 RSS 最大的 10 项映射。
客户端应调用 get_process_memory(pid=1234, mapping_limit=10)。进程不存在或无权限时,调用应明确失败而不是返回伪造结果。
3. 调度算法对比
对任务 A(到达0、运行4) 和 B(到达0、运行2) 使用时间片1的 RR 调度,解释等待时间。
客户端应调用 simulate_cpu_scheduling。正确结果中 A、B 的累计等待时间都为 2。
更多示例见 docs/demo.md。
测试
ruff check .
pytest --cov=openeuler_mcp --cov-report=term-missing测试覆盖页面置换、RR 累计等待时间、磁盘分配回滚、路径边界、当前系统服务,以及 MCP Server 的 12 个工具注册、Schema 和结构化调用。
评测
evaluation/cases.jsonl 包含 20 条任务,其中两条用于验证模型不会选择不存在的危险写工具。把客户端调用轨迹整理为 result.example.jsonl 的格式后运行:
python evaluation/evaluate_results.py evaluation/result.example.jsonl输出工具选择正确率、参数有效率、调用成功率和任务完成率;各项指标根据输入的客户端调用轨迹计算。
文档
已知限制
仅提供本地 stdio transport,不包含远程 HTTP 和认证。
文件变化通过元数据轮询观察,快速发生并恢复的变化可能被漏掉。
进程和系统状态具有瞬时性,采集期间进程可能退出或权限可能变化。
磁盘分配和 CPU/页面调度仅计算模拟结果,不会修改真实操作系统状态。
License
Available Tools
12 toolsanalyze_file_distributionC
Summarize file counts, sizes, extensions, and largest paths in a directory.
Inputs: path, top_n, max_depth, max_files. Returns: root, max_depth, extensions, largest_files, +3 more
get_filesystem_infoB
Read mounted filesystems, space usage, and inode counts when available.
Inputs: no named parameters. Returns: partitions, captured_at
get_memory_infoA
Read current system RAM, swap, and Linux /proc/meminfo values.
Inputs: no named parameters. Returns: swap, memory, source, captured_at, +1 more
get_process_memoryA
Read RSS, VMS, and the largest memory mappings for one process.
Inputs: pid (required), mapping_limit. Returns: pid, name, source, mappings, +5 more
get_process_treeB
Read a bounded parent-child process tree from a selected root PID.
Inputs: root_pid, max_depth, max_nodes. Returns: tree, root_pid, max_depth, truncated, +2 more
monitor_context_switchesB
Sample system-wide or per-process context-switch deltas.
Inputs: pid, duration_seconds, interval_seconds. Returns: pid, scope, samples, duration_seconds, +1 more
monitor_file_metadataB
Poll file size and timestamps; this is metadata polling, not access tracing.
Inputs: path (required), duration_seconds, interval_seconds. Returns: path, events, method, duration_seconds, +1 more
sample_cpu_time_ratiosB
Sample CPU time deltas and normalize them into per-state ratios.
Inputs: duration_seconds, interval_seconds. Returns: samples, duration_seconds, interval_seconds
sample_memory_trendC
Sample system memory usage and classify its short-term direction.
Inputs: duration_seconds, interval_seconds. Returns: trend, samples, change_ratio, duration_seconds, +1 more
simulate_cpu_schedulingC
Simulate CPU scheduling and report waiting, turnaround, and response metrics.
Inputs: jobs (required), algorithm, time_slice. Returns: jobs, metrics, timeline, algorithm, +1 more
simulate_disk_allocationB
Simulate transactional contiguous, linked, or indexed block allocation.
Inputs: scenario (required), strategy, block_size_bytes. Returns: strategy, allocations, free_blocks, total_blocks, +5 more
simulate_page_replacementC
Simulate FIFO, LRU, CLOCK, or optimal page replacement.
Inputs: reference_string (required), algorithm, frame_count. Returns: hits, steps, faults, hit_rate, +4 more
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 12 tools
Tools cover distinct areas like memory, filesystem, processes, and CPU scheduling, but some memory-related tools (get_memory_info, sample_memory_trend, get_process_memory) could be confused. Overall, each tool has a specific purpose with limited overlap.
Names use mixed verb patterns (get_, monitor_, sample_, simulate_, analyze_) rather than a consistent verb_noun structure. While readable, the lack of pattern may cause hesitation in selecting the correct tool.
With 12 tools, the count falls well within the ideal 3-15 range for a system toolkit. Each tool serves a clear function without unnecessary bloat or sparseness.
The toolkit covers memory, filesystem, process, and CPU scheduling domains, but lacks basic tools like listing all processes, CPU info, or network stats. Notable gaps exist for a comprehensive system toolkit.
Related MCP Connectors
An MCP server that provides read access to your cloud storage providers, bank accounts and more.
Read-only MCP server for the OPERANT AI operating-agent calibration benchmark.
Linux server administration over OAuth MCP: shell, Docker, systemd, files and monitoring.
A free, read-only remote MCP server with templates for a file-based business operating system.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceA read-only MCP server for Linux and macOS system administration, diagnostics, and troubleshooting, supporting remote SSH execution and multi-host management.Apache 2.0
- AlicenseBqualityDmaintenanceA powerful MCP server for reading and modifying Linux /proc filesystem values, providing system monitoring, process management, and sysctl operations via JSON-RPC and SSE.14MIT
- AlicenseAqualityBmaintenanceA Linux system monitoring MCP server that provides real-time information on CPU, memory, disk, network, processes, Docker, security, and more via MCP tools.10653MIT
- -licenseNot gradedqualityNot gradedmaintenanceA read-only system observability and OS algorithm lab MCP server for openEuler/Linux, encapsulating memory, filesystem, process, and CPU info into typed tools for reliable LLM client use.-