reddit · SaaS · 10-17 13:50 50 分 ← 返回列表

I got tired of researching which skills and MCP servers to give my coding agent, so I built a tool that picks them from your project's stack (free tier)

https://preview.redd.it/fz9uf8i801wh1.png?width=2902&format=png&auto=webp&s=a33b81b80300d5ca98e95921161712b02623eb88 Every coding agent (Claude Code, Codex, Kimi, Gemini CLI…) has the same two problems: Its training data stops at a cutoff, so it confidently uses APIs that were renamed or removed since. There are thousands of skills and MCP servers out there, and figuring out which ones are worth adding (and which ones are safe) is a research project of its own. C

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：编码 agent 需要人工调研适配项目技术栈的 skills 和 MCP servers，且模型训练数据过期会带来 API 幻觉、破坏性变更和安全风险。

目标用户：使用 Claude Code、Codex、Kimi、Gemini CLI 等 coding agent 的开发者与工程团队。

证据：I got tired of researching which skills and MCP servers to give my coding agent, so I built a tool that picks them from your project's stack. Tasks solved: 38% → 83% on the free tier. Cost per solved task: $0.112 → $0.071.

变现：Freemium：免费层限量；Pro 订阅按开发者收费，提供私有仓库依赖分析、持续安全扫描、团队策略与审计、优先更新源；Team 版按席位收费。

MVP：2 周内做一个 MCP server：读取 package.json/requirements 等依赖名和版本，匹配人工精选的 skills/MCP 清单，展示训练截止后的破坏性变更和基础安全标记，不上传源码。

竞品：Context7、Smithery、mcp.so、Glama MCP

摘要：作者因厌倦为 coding agent 调研 skills 和 MCP servers，做了 CodexGuild，读取项目依赖名和版本，推荐适配且安全扫描过的 skills/MCP，并提示模型训练截止后的 breaking changes。自测同一 Claude Code+Sonnet 在 20 个真实任务上解决率从 38% 提升到 83%，每任务成本从 $0.112 降到 $0.071，提供免费层。帖子是产品发布，未展示外部用户付费意愿或收入数据。