Let Claude Code users set the docs MCP server URL with VAADIN_DOCS_MCP_URL

未关闭
#3 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
35/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
技术栈
json, markdown, python

调研方向

The server is declared in .mcp.json and the Claude-only declaration would go in .claude-plugin/plugin.json. The URL check lives in scripts/check_plugin.py, which needs to accept the ${VAADIN_DOCS_MCP_URL:-https://...} form, and the user-facing steps are in plugin/README.md. Done means both manifests resolve to the same URL once the default is substituted, the check script passes, and the README describes the variable. The release review against the Claude plugin directory is a separate maintainer step.

由索引模型根据 Issue 内容生成。

描述

enhancement

Customers on air-gapped networks may run their own Vaadin documentation MCP server. The plugin's .mcp.json hard-codes https://mcp.vaadin.com/docs, so today they have to block the hosted server and add their own by hand.

Proposal

Let Claude Code users point the vaadin server at their own URL with an environment variable that defaults to the hosted one:

"url": "${VAADIN_DOCS_MCP_URL:-https://mcp.vaadin.com/docs}"

An admin could set it for a whole organization through env in managed settings. The server keeps its name and its tools, so the skills don't change.

What I checked (Claude Code 2.1.318, Codex 0.163)

  • Claude Code expands ${VAR:-default} in a plugin MCP server's url. It reads the value from the shell and from env in settings.
  • Codex reads the same .mcp.json and doesn't expand variables, so it would use the URL literally and the server would fail. In Codex's source (0.167), only ${PLUGIN_ROOT} and ${PLUGIN_DATA} are expanded, and only for stdio servers. Codex plugins also have no settings mechanism.
  • So the variable can't go in the shared .mcp.json. What works is a Claude-only declaration under mcpServers in .claude-plugin/plugin.json. Claude Code loads it together with .mcp.json, and on a name clash it used the inline server. Codex reads only .mcp.json, through its own manifest.
  • claude plugin validate --profile directory raised nothing about the variable in the URL.

To do

  • Declare the vaadin server inline in .claude-plugin/plugin.json with the variable in its URL. Leave .mcp.json as it is for Codex.
  • Extend scripts/check_plugin.py so the two declarations can't drift: the same URL once the variable's default is substituted. Its https check needs to accept ${VAADIN_DOCS_MCP_URL:-https://…}.
  • Document the variable in plugin/README.md, replacing the block-and-add steps for Claude Code.
  • Before release, check that the Claude plugin directory's review accepts the variable in the URL, and that installs from claude.ai and Cowork still reach the hosted server.

Notes

  • That the inline server wins over the same-named one in .mcp.json is what Claude Code did in testing. I didn't find it documented.
  • A VAADIN_DIRECTORY_MCP_URL would only be useful if a self-hosted Directory server exists.
  • I left out Claude Code's userConfig (prompted plugin settings). It asks users for the value when they enable the plugin, and the docs don't confirm that an admin can fill it in for them.
  • Codex users would keep replacing the server in config.toml: a [mcp_servers.vaadin] entry takes the place of the plugin's server of the same name. That works in 0.163 but OpenAI doesn't document it.
主要语言
Go
星标
0
派生
0
平均合并
4 天 10 小时
30 天内合并 PR
3

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

相似的 Issue

更多 Go Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。