mcp-check-fixture
Minimal MCP server for testing GitHub PR checks pipeline, with deterministic tools echo, add, and server_info.
README
mcp-check-fixture
The MCP server MCPJam's GitHub PR checks are validated against. This directory is
the seed for the standalone public repository mcpjam/mcp-check-fixture —
the only repo the walking-skeleton check pipeline is wired to
(server/services/github-checks/recipes.ts).
It lives here so it is reviewed alongside the pipeline that depends on it, and so the ops step is "copy and push" rather than "write from scratch".
Publishing it
cp -R test-servers/mcp-check-fixture /tmp/mcp-check-fixture
cd /tmp/mcp-check-fixture
git init && git add -A && git commit -m "mcp-check-fixture: minimal MCP server"
gh repo create mcpjam/mcp-check-fixture --public --source=. --push
package-lock.json is committed here on purpose, and must stay committed in the
published repo: the run recipe uses npm ci, which fails outright without one.
A repo missing its lockfile reports build_failed — correctly, but confusingly.
@types/node is likewise a real devDependency, not an assumption about the
build environment: tsconfig.json declares "types": ["node"], and this package
is standalone with no workspace to hoist it from.
The two things that must not be "cleaned up"
It binds 0.0.0.0, not 127.0.0.1. E2B's host bridge only forwards to ports
bound on all interfaces. A loopback-only server is invisible from outside the
sandbox and the check reports server_unhealthy with a stderr tail that may not
mention the bind address at all.
It answers initialize on /mcp immediately. That handshake is the health
probe — the worker polls it until it succeeds. Anything that delays it (a
warm-up, a lazily mounted route) reads as an unhealthy server.
Run recipe
The recipe hardcoded in the Inspector for this repo:
| field | value |
|---|---|
build |
npm ci && npm run build |
start |
npm start |
port |
3001 |
mcpPath |
/mcp |
Changing any of those means changing recipes.ts in the same breath.
Tools
Three, all deterministic and side-effect-free — the eval suite's job is to prove the pipeline works, not to be an interesting test:
echo({ text })→ the same textadd({ a, b })→ the sumserver_info()→{"name":"mcp-check-fixture","version":"1.0.0"}
Locally
npm install && npm run build && npm start
curl -s localhost:3001/mcp -H 'content-type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}'
Exercising the check's failure modes
Each of these is a one-line change, and each should produce a specific check conclusion — useful for verifying the pipeline end to end:
| Change | Expected outcome | Check conclusion |
|---|---|---|
| break a tool's return value | evals_failed |
failure |
| introduce a type error | build_failed |
failure |
throw at the top of server.ts |
server_unhealthy |
failure |
bind 127.0.0.1 instead of 0.0.0.0 |
server_unhealthy |
failure |
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。