dsh-mcp-lens

Every time an AI agent loads a large MCP tool set, the full schema for each tool is packed into the prompt, wasting tokens, slowing responses, and inflating the bill. dsh-mcp-lens tackles this schema overhead by routing 1,000+ tools through just two progressive-disclosure interfaces, so the model only sees the descriptions it actually needs at that moment. Project benchmarks report up to 90% less data transfer, roughly 80% lower token usage, and 50% faster response times. Beyond routing, it ships a local calculator that keeps simple math off the AI bill entirely, and a CI Budget Action that tracks usage in real time, warns you as you approach your spending ceiling, and helps forecast future costs. Everything runs locally on Windows 10/11 and plugs into the MCP standard and DeepSeek Harness with no external server required.

The plugin is a natural fit for three audiences: solo developers and small teams juggling dozens or hundreds of MCP tools where schema bloat is already a bottleneck; researchers and students running heavy AI workloads on a fixed budget; and freelancers or SMEs who need per-token spend to stay predictable. If you are already working inside DeepSeek Harness with a growing tool list, dropping dsh-mcp-lens into your pipeline should make a visible difference in your token bill within the first session.