I built MCP Context Budget after seeing coding-agent sessions lose useful context before the actual task even started. The problem is subtle: connect enough MCP servers and the tool definitions alone can consume thousands of tokens. Then tool responses add another layer of context pressure. By the time the agent starts working, part of the context window is already gone. This tool treats MCP context usage like build infrastructure: * scan MCP tool definitions * estimate schema-token and response-token cost separately * select a smaller task-relevant tool set * write a lockfile * fail CI when the budget regresses It is local-first, MIT licensed, and has a dependency-free core. No hosted service, no proxy, nothing leaves your machine unless you explicitly use optional local Ollama ranking. The goal is simple: make context bloat visible in review, not mid-session.

MCP Context Budget

Measure and cap MCP context bloat before agents run

Measure and cap MCP context bloat before agents run