allure-testops-mcp
allure-testops-mcp
Allure TestOps 的 MCP 服务器。允许 LLM 智能体(Claude Code、Cursor、OpenCode 等)通过 Allure REST API 查询项目、测试运行(launches)、测试用例和测试结果。
基于 Python、FastMCP 和 stdio 传输协议。
适用于任何 Allure TestOps 实例 — SaaS 版 qameta.io 或自托管/本地部署版。专为企业网络设计:支持配置代理绕过、可选的 SSL 验证开关以及 API 令牌认证。
设计亮点
工具注解 — 每个工具都标记为
readOnlyHint: True/openWorldHint: True。所有 6 个工具均为只读;MCP 客户端不会要求确认。每个工具均提供结构化输出 — 每个工具都声明了
TypedDict返回类型,因此 FastMCP 会自动生成outputSchema,且每个结果都包含structuredContent(类型化负载)和预渲染的 Markdown 文本块。结构化错误处理 — 将认证、404、403、429、5xx 和缺少环境变量等错误转换为可操作的消息(例如:“认证失败 — 请验证 ALLURE_TOKEN 是否具有 API 权限”)。
Pydantic 输入验证 — 每个参数都有类型约束(范围、长度、字面量),并自动公开为 JSON Schema。
分页 — 列表工具返回一个包含
page、total、has_more和next_page的pagination块。通过 MCP 上下文报告进度 — 执行多次 API 调用的工具(
allure_get_project_statistics、allure_search_failed_tests)以及allure_list_test_cases会发出ctx.report_progress+ctx.info事件,以便兼容的客户端渲染进度条和步骤标签。
Related MCP server: Allure TestOps MCP
功能
涵盖日常 Allure TestOps 工作流的 6 个工具:
发现
allure_list_projects— 获取所有项目(包含 ID、名称、缩写)allure_get_project_statistics— 获取测试用例计数、自动化率、最近一次运行摘要
运行与结果
allure_list_launches— 获取包含通过/失败统计信息的近期运行记录allure_get_test_results— 获取某次运行中的测试结果(可按状态过滤)allure_search_failed_tests— 获取最近一次或指定运行中 FAILED/BROKEN 的测试
测试用例
allure_list_test_cases— 获取测试用例,支持自动化/手动过滤(每个结果还包含其层级,例如UNIT/API/E2E)
安装
需要 Python 3.10+。
# via uvx (recommended)
uvx --from allure-testops-mcp allure-testops-mcp
# or via pipx
pipx install allure-testops-mcp配置
简短版本 — claude mcp add:
claude mcp add allure -s project \
--env ALLURE_URL=https://allure.example.com \
--env ALLURE_TOKEN=your-api-token \
--env ALLURE_SSL_VERIFY=true \
-- uvx --from allure-testops-mcp allure-testops-mcp或者在 ~/.claude.json / 项目 .mcp.json 中配置:
{
"mcpServers": {
"allure": {
"type": "stdio",
"command": "uvx",
"args": ["--from", "allure-testops-mcp", "allure-testops-mcp"],
"env": {
"ALLURE_URL": "https://allure.example.com",
"ALLURE_TOKEN": "${ALLURE_TOKEN}",
"ALLURE_SSL_VERIFY": "true"
}
}
}
}查看 .env.example 获取所有支持的环境变量模板。
检查:
claude mcp list
# allure: uvx --from allure-testops-mcp allure-testops-mcp - ✓ Connected环境变量
变量 | 必需 | 描述 |
| 是 | Allure TestOps URL (例如 |
| 是 | Allure TestOps 的 API 令牌 (个人资料 → API 令牌) |
| 否 |
|
使用示例
在 Claude Code 中:
“列出所有 Allure 项目”
“显示项目 63 的最近 10 次运行”
“项目 175 最近一次运行中失败的测试”
“项目 842 的自动化率”
“运行 12345 中状态为 FAILED 的测试结果”
安全考量
API 令牌从
ALLURE_TOKEN环境变量读取 — 绝不会在命令行中传递,也不会写入日志。机密信息不会在工具响应中回显(没有
stat.request_headers转储,没有session.auth反射)。自签名 SSL 可通过
ALLURE_SSL_VERIFY=false选择性开启 — 默认为true。在公共网络上禁用验证存在安全风险;仅在受信任的企业实例中使用。禁用代理发现 (
session.trust_env = False) — MCP 会刻意忽略HTTP_PROXY/HTTPS_PROXY环境变量,因此会话不会被静默路由到非预期的代理。如果您的 Allure 实例只能通过代理访问,请在requests可以直接解析的环境中运行 MCP。未公开写操作 — 所有 6 个工具均为只读。即使 API 令牌具有写权限,此 MCP 服务器也无法在 Allure TestOps 中创建、修改或删除任何内容。
通过 Pydantic 进行输入验证 — 每个工具参数都有类型和边界限制(ID 必须 ≥ 1,分页上限为 200-500)。
速率限制
Allure TestOps 强制执行实例级速率限制(API 令牌通常约为每分钟 60 次请求)。当遇到 HTTP 429 时,MCP 会返回一个可操作的错误,建议您:
等待 30-60 秒后再重试。
减小
size参数(测试结果默认为 50,项目默认为 200)。使用更小的页面大小进行分页。
有两个工具在内部执行多次 API 调用:
allure_get_project_statistics— 3 次调用(TC 计数 + 运行记录 + 运行统计)。allure_search_failed_tests— 2-3 次调用(解析最新运行 + FAILED + BROKEN)。
两者都使用 MCP Context 来报告每一步的进度;allure_list_test_cases 也会发出单个进度事件。请在兼容的客户端中监控进度流。
开发
git clone https://github.com/mshegolev/allure-testops-mcp.git
cd allure-testops-mcp
pip install -e '.[dev]'
pytest直接运行服务器(stdio 传输,等待 stdin 的 MCP 消息):
ALLURE_URL=... ALLURE_TOKEN=... allure-testops-mcp许可证
MIT © Mikhail Shchegolev
Available Tools
11 toolsallure_get_project_statisticsARead-onlyIdempotent
Get summary statistics for an Allure project.
Returns TC count, automation rate, and the last closed launch's pass/fail breakdown. Performs 3-4 API calls — progress is reported via MCP Context (visible as progress updates in compatible clients).
Args:
project_id: Allure project ID (see allure_list_projects).
ctx: MCP Context injected by FastMCP (used for progress reporting;
never supplied by the agent directly).
Returns:
dict with keys:
- project_id (int)
- total_test_cases (int)
- automated_test_cases (int)
- manual_test_cases (int)
- automation_rate_pct (float)
- last_launch_id (int | None): latest closed launch
- last_launch_name (str | None)
- last_launch_passed / last_launch_failed / last_launch_broken (int)
- last_launch_total (int)
- recent_launches_count (int): launches examined to find the latest closed one
Examples:
- "How automated is project 63?" -> project_id=63, read automation_rate_pct
- "What was the last passing run for project 175?" -> read last_launch_passed
Don't use when:
- You need per-test detail (use ``allure_get_test_results``).
- You need the full launch history (use ``allure_list_launches``).
Inputs: project_id (required). Returns: project_id, total_test_cases, automated_test_cases, manual_test_cases, +8 more
allure_get_test_caseARead-onlyIdempotent
Get one test case's full detail — fields, status/layer, tags, and steps.
Unlike allure_list_test_cases (summaries), this returns the body of a
single test case: description, precondition, expected result, and the
manual scenario steps (flattened with a depth marker). Use it to read
or review the actual content of a test case.
Returns:
dict with id, name, project_id, automated,
description, precondition, expected_result, status,
layer, tags and steps (each: depth, keyword,
name, expected_result). steps is empty when
include_scenario is false or the case has none.
Inputs: test_case_id (required), include_scenario. Returns: id, name, project_id, automated, +9 more
allure_get_test_case_custom_fieldsARead-onlyIdempotent
List the custom-field values set on a test case.
Returns each value flattened to field_id / field_name (the custom
field) and value_id / value_name (the chosen value) — e.g. field
"Priority" → value "High". These are not included in
allure_get_test_case; fetch them here when you need a test case's
custom-field assignments.
Inputs: test_case_id (required). Returns: test_case_id, count, custom_fields
allure_get_test_resultsARead-onlyIdempotent
Get individual test results inside a launch, optionally filtered by status.
Args:
launch_id: Allure launch ID (from allure_list_launches).
status: Filter — PASSED / FAILED / BROKEN / SKIPPED. None returns all.
page: 0-based page index.
size: Items per page (1-200; default 50).
Returns:
dict with keys:
- launch_id (int)
- count (int)
- pagination (dict)
- results (list): each with id / name / status /
duration_ms / error (first 300 chars of statusMessage)
Examples:
- "FAILED tests in launch 12345" -> launch_id=12345, status="FAILED"
- "All results in launch X, second page" -> launch_id=X, page=1
Don't use when:
- You want only FAILED+BROKEN (``allure_search_failed_tests`` does both in one call).
Inputs: launch_id (required), status, page, size. Returns: launch_id, count, pagination, results
allure_list_custom_fieldsARead-onlyIdempotent
List the custom fields defined on a project (its schema).
Returns each field's field_id, name, single_select and
required. Use it to discover which custom fields a project has before
reading a test case's values with allure_get_test_case_custom_fields.
Built-in metadata fields (Epic/Feature/Story/Component/Suite) use negative
ids; project-specific custom fields use positive ids.
Inputs: project_id (required). Returns: project_id, count, custom_fields
allure_list_launchesARead-onlyIdempotent
List recent launches for a project, newest first.
Each launch carries a pass/fail/broken/skipped breakdown from Allure's
statistic field. Pagination info is returned in the pagination
block (use next_page to continue).
Args: project_id: Allure project ID. page: 0-based page index. size: Items per page (1-100; 20 is usually enough for triage).
Returns:
dict with keys:
- project_id (int)
- count (int): items in this response
- pagination (dict): page / size / total /
total_pages / has_more / next_page
- launches (list): each with id / name / status /
created_date / passed / failed / broken /
skipped / total
Examples:
- "Last 10 launches for project 63" -> project_id=63, size=10
- "Older launches beyond page 1" -> repeat with page=1
Don't use when:
- You need test results inside a launch (``allure_get_test_results``).
- You need just the latest FAILED/BROKEN tests (``allure_search_failed_tests``).
Inputs: project_id (required), page, size. Returns: project_id, count, pagination, launches
allure_list_layersARead-onlyIdempotent
List the test layers defined in a project (id, name).
Use this to discover valid layer names/ids before setting a layer on
allure_create_test_case / allure_update_test_case. Built-in
layers use negative ids (e.g. API Tests = -3).
Returns:
dict with project_id, count and layers (each: id, name).
Inputs: project_id (required). Returns: project_id, count, layers
allure_list_projectsARead-onlyIdempotent
List all projects in the Allure TestOps instance.
Use this first to discover which project IDs exist — all other tools
take a project_id that you can look up here.
Returns:
dict with keys:
- count (int): number of projects in this response
- projects (list): each item has id, name, abbreviation
Examples:
- "Which projects exist in Allure?" -> default call, take the names/ids
- "Find project by abbreviation" -> iterate projects and match
Don't use when:
- You already know the project id (skip discovery, go straight to the target tool).
Inputs: page, size. Returns: count, projects
allure_list_statusesARead-onlyIdempotent
List the test-case statuses defined in a project (id, name, color).
Use this to discover valid status names/ids before setting a status on
allure_create_test_case / allure_update_test_case. Built-in
statuses use negative ids (e.g. Draft = -1).
Returns:
dict with project_id, count and statuses (each: id,
name, color).
Inputs: project_id (required). Returns: project_id, count, statuses
allure_list_test_casesARead-onlyIdempotent
List test cases for a project with optional automation and ownership filters.
Each TC carries id, name, automated, status, layer
(e.g. UNIT, API, E2E), the createdBy / lastModifiedBy
audit usernames, and a flat list of tag names. Caveat: the audit
fields and tags are only populated when owner is set, because
Allure's plain /testcase endpoint returns a compact projection
that omits them — the owner path uses __search which returns
the full projection.
Args:
project_id: Allure project ID.
ctx: MCP Context (auto-injected by FastMCP for progress reporting).
automated: True — only automated, False — only manual,
None — both.
owner: Optional Allure username. When set, the response is narrowed
to TCs where createdBy = owner OR lastModifiedBy = owner
(case-sensitive, exact match), enforced server-side via
Allure's RQL __search endpoint. The username must match
[A-Za-z0-9._@-]+ — anything else is rejected at the MCP
input layer (Pydantic pattern) to prevent RQL injection.
**Why "creator/modifier" and not "owner".** Allure TestOps does
not expose a separate ``owner`` field in RQL on most
deployments — the closest stable proxy for "TCs I touched" is
the union of ``createdBy`` and ``lastModifiedBy``.
**Trade-off when ``owner`` is set.** ``__search`` does not
accept the ``automated`` query parameter, so an ``automated``
filter combined with ``owner`` is applied **client-side after
the page is fetched**. ``pagination`` then reflects the raw
owner-filtered set, not the further automation-filtered view —
a fetched page of 50 may shrink. Raise ``size`` (max 200) or
iterate ``page`` for full coverage.
page: 0-based page index.
size: Items per page (1-200; default 50).Returns:
dict with keys:
- project_id (int)
- count (int): items in this response (post any client-side automated filter)
- pagination (dict): raw Allure paging
- test_cases (list): each item carries id / name /
automated / status / layer / created_by /
last_modified_by / tags
Examples:
- "How many manual TCs does project 63 have?" -> project_id=63, automated=False, read pagination.total
- "First 200 automated TCs" -> automated=True, size=200
- "My manual TCs in project 63" -> project_id=63, automated=False, owner="jdoe", size=200
Don't use when:
- You need just the automation % (``allure_get_project_statistics``).
Inputs: project_id (required), automated, owner, page, +1 more. Returns: project_id, count, pagination, test_cases
allure_search_failed_testsARead-onlyIdempotent
Find FAILED and BROKEN tests in the most recent (or given) launch.
Useful for triage: "what's broken in the latest run" without listing every test. Performs up to 3 API calls; progress reported via MCP Context.
Args:
project_id: Allure project ID.
launch_id: Specific launch ID. If None, the latest launch is used.
limit: Max failures per status (so up to 2 * limit total).
ctx: MCP Context (auto-injected).
Returns:
dict with keys:
- launch_id (int): the resolved launch (latest if not passed in)
- failed_count (int)
- results (list): id / name / status / duration_ms / error
Examples:
- "What's failing in project 63 right now?" -> project_id=63
- "Failures in launch 98765" -> project_id=N, launch_id=98765
Don't use when:
- You need PASSED tests too (use ``allure_get_test_results`` without status filter).
Inputs: project_id (required), launch_id, limit. Returns: launch_id, failed_count, results, reason
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 11 tools
Each tool targets a distinct entity or operation (projects, statistics, test cases, launches, results, failures, custom fields, layers, statuses) with clear boundaries and explicit 'Don't use when' guidance to prevent misselection.
All tools follow a consistent 'allure_verb_noun' pattern (e.g., allure_list_projects, allure_get_test_case), making the naming predictable and easy to navigate.
With 11 tools, the set is well-scoped for a test management server, covering essential query and list operations without being overwhelming.
The tools are heavily read-focused, missing create/update/delete operations for test cases and other entities, which are fundamental to full lifecycle management. The descriptions reference 'allure_create_test_case' but it is not included.
Maintenance
Related MCP Connectors
Official MCP server for Qase — manage test cases, runs, suites, defects via AI tools.
Run, debug, and triage tests from your IDE using natural language, no dashboard switching, no manual data transfers. The TestMu AI (formerly LambdaTest) MCP Server is a single remote server exposing four tool suites: HyperExecute — analyze your project, generate YAML configs and test runner commands, then monitor jobs and sessions. Automation — pull a TestID's details plus command, network, and console logs into one chat for instant root-cause analysis. Includes mobile app upload. SmartUI — explain pixel, layout, DOM, and perceptual changes in a visual regression run, with context-aware React/HTML/CSS fixes. Accessibility — audit any public URL or a local React app against WCAG and get ready-to-apply remediation steps. Connects over https://mcp.lambdatest.com/mcp using OAuth 2.1 — no API keys in your config. One-click install in Cursor; works with Claude, GitHub Copilot, Cline, and any MCP client. Tests execute on the TestMu AI cloud: 3,000+ browsers and 10,000+ real devices.
MCP server for building and testing AI agents with multi-model experimentation and insights.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Related MCP Servers
- AlicenseBqualityAmaintenanceAI-native Model Context Protocol (MCP) server for TestRail. Lets Claude, Cursor, Windsurf, and other AI assistants browse projects, create and update test cases, kick off test runs, and record results through natural-language conversation — with strongly-typed tool schemas and per-project custom field validation that helps LLMs generate valid TestRail requests on the first try.271,151 npm45Apache 2.0
- AlicenseCqualityCmaintenanceProduction-ready MCP server for Allure TestOps, enabling test case, launch, test result, and test plan management via natural language.100297 npm2MIT
- AlicenseBqualityBmaintenanceAn MCP server for Allure TestOps that enables AI agents to manage test cases, defects, test plans, and other test management entities with clear, easy-to-use tools and helpful error guidance.70408 PyPI8Apache 2.0
- AlicenseAqualityDmaintenanceAn MCP server that connects AI agents to TestOps, enabling test case management, launches, defects, and analytics through natural language.48MIT