RU

MCPtoCodeMode hides all MCP servers behind three tools and a sandbox

Published: 2026-10-23 · Author: AI Release · @ai_release1
MCPtoCodeMode hides all MCP servers behind three tools and a sandbox

⚡ The gist in 5 seconds - The minkaworks/mcptocodemode project puts all MCP servers behind three tools and a sandbox. - Compatibility is claimed with Codex, Claude Code, Cursor, Grok, OpenCode, Pi, and any MCP client. - The names of the three tools, the sandbox composition, and installation commands are not disclosed in the repository description. ### 🔍 What was found On GitHub, a repository called minkaworks/mcptocodemode has appeared — an open-source project that, according to the author's description, solves the problem of an agent's prompt getting "cluttered" with MCP tool definitions. The wording from the repository is direct: "Stop flooding your agent's prompt with MCP tools." The list of supported clients includes Codex, Claude Code, Cursor, Grok, OpenCode, Pi, as well as any MCP client — meaning the implementation is positioned as a universal wrapper rather than a plugin for a single environment. The mechanics are described briefly: each MCP server is placed behind three callable tools and a sandbox. Instead of passing the agent the full list of tools from all connected servers, the client sees a compact interface of three entry points, while the MCP servers themselves run in an isolated environment. There is no more detail in the source: the names of the three tools, the isolation mechanism, the license, the dependency list, and setup instructions are not provided in the repository description, so implementation details remain unknown for now. ### 💡 Why it matters The problem the author describes is familiar to anyone who has built agent setups: every connected MCP server adds its tool schemas to the context, and with a dozen servers, a noticeable chunk of the window is spent on service descriptions instead of the task. Reducing servers to three tools plus a sandbox is an attempt to cut that overhead without abandoning MCP as a protocol. Whether the approach works in practice, whether token-savings metrics will appear, and when the project will move past its early stage cannot be determined from the repository description.

🤖 AI summary
#MCP#AI-агенты#open#Codex#Claude#AI#LLM#нейросети
← PreviousiPhone Duo is here: Apple folds late, ships short

Source: github.com · post in Telegram