An MCP server that MCP clients can connect to (read-only first)

未关闭
#1,607 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
一周以上
新手友好度
18/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
技术栈
python
领域
api, backend, devtools

调研方向

Start with docs/mcp-server.md and the manifest served at GET /api/v1/mcp/manifest, then the REST endpoints under /api/v1/ai/. The issue is design-gated: the language, package, authentication and tool list must go through plan and review before any code. Done means a plan approved by maintainers, then a stdio server that reads flags and experiments with the caller's credentials.

由索引模型根据 Issue 内容生成。

描述

enhancement

Today the platform serves an MCP manifest at GET /api/v1/mcp/manifest (five tools described) and REST endpoints under /api/v1/ai/, but it does not run an MCP server: an MCP client such as Claude Code or Cursor cannot connect to it, as docs/mcp-server.md says. Six of the eight competitors in the 31 October research have one (GrowthBook on every tier, LaunchDarkly, Datadog, PostHog among them).

Wanted before launch (founder, 2026-11-02): a real MCP server that an MCP client can connect to.

  • First version read-only, over the existing API, with the caller's own credentials, so it can do nothing the caller's role cannot: list and read flags, evaluate a flag for a user, list experiments and read their results. Write tools (toggle a flag, create an experiment) come after, under the same role checks as the API.
  • Runs locally next to the client (stdio), pointed at the user's own Controlvar instance; no hosted service.
  • A small tool set, each tool's description and inputs pinned by a test, and the existing manifest either replaced by it or kept true.
  • Docs: docs/mcp-server.md rewritten to say how to connect a client.

Publishing the package to a registry is a separate step that waits for the founder at that moment.

The design (language, package, authentication, tool list) goes through the usual plan and review before any code.

主要语言
Python
星标
0
派生
4
平均合并
5 小时 36 分钟
30 天内合并 PR
509

环境准备

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。

Controlvar/controlvar 的其他 Issue

查看 Controlvar/controlvar 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。