AI AgentsDevToolsSecurityMCP

MCP Trim—Schema thinning and firewall for MCP

Control and slim down MCP tool calls for Claude and Codex agents

Problem
As the MCP ecosystem explodes, AI coding agents mount more and more tools, causing context bloat, token waste, and risky tool calls. Someone built CtxWise as a schema tree-shaker and firewall, proving this is a fresh, real pain point, but the tooling is still immature.
Solution
A middleware that sits between the agent and MCP tools: strips unused schema fields to save tokens, enforces allow/deny rules and approval flows for sensitive tools (file deletion, network, shell), and logs every call.
Users
Heavy developers running multiple AI coding agents, AI toolchain teams, and tech leads concerned about agent safety
MVP
MCP proxy gateway that intercepts and trims tool schemas; allow/deny rules by tool name or argument regex; a local log viewer; integrations for both Claude Code and Codex; rules managed via YAML config
Revenue
Open-source core with a team subscription at $10/user/month for rule sharing, audit log export, and SSO; or a hosted gateway billed per call
Why now
MCP went mainstream in 2025-2026, but tool governance and security are nearly blank. Indie developers know their own agent workflows best and can validate rule sets quickly.

Source discussions · 2

  1. Hacker NewsAsk HN: Has anyone decompiled Claude Code yet?1↑ 8
  2. Hacker NewsShow HN: CtxWise – MCP schema tree-shaker and firewall for Claude and Codex1↑ 0
← Back to list