Jill, an indie app builder and self-described vibe coder, has shipped over 30 small business applications—invoicing systems, inventory trackers, packing slip generators, tax calculators. But lately she's been working on something unexpected: an open standard for AI agent interoperability called Blueprint Protocol.

The Visibility Problem

"The problem I kept running into: even the best AI agents hallucinate when they look at a web app," Jill writes in her DEV.to post. "They guess where to click. They call the wrong tools. They fail silently and confidently." With 30 apps that technically existed, she discovered that no agent could actually see them—or more precisely, none of them had any reliable way to be seen.

How Blueprint Protocol Works

The solution is deceptively simple: a blueprint.txt file placed at the root of your application. This file tells agents exactly what your app can do, which tools to call, what's human-only, and how to complete each workflow step by step. The key insight? No MCP (Model Context Protocol) server required. It works for any agent trying to navigate any web app. For environments that do use MCP specifically, the blueprint eliminates an entire discovery round trip. Without one, an agent calls tools/list first, reads through all N tool definitions, picks one, then calls it. With a blueprint, the agent already knows the tool name and exact parameters—it calls tools/call directly. That sounds incremental until you realize it's the removal of a whole network hop.

The Benchmark Data

Jill ran a controlled test across four models—Claude, GPT-4o, Gemini, and Grok—using five candidate MCP servers with one correct match. Agents were instructed to find the best server efficiently, with no blueprint references in the harness instructions. Discovery was organic. | Model | Without Blueprint | With Blueprint | Reduction | |-------|------------------|----------------|-----------| | Claude | 58 pts | 13 pts | ~78% | | GPT-4o | 58 pts | 28 pts | ~54% | | Gemini | 23 pts | 9 pts | ~61% | | Grok | 58 pts | 58 pts | 0% | Overhead was scored by weighted resource inspection cost—LLMs.txt at 10 points, blueprint.txt at 1 point, reflecting typical real-world document sizes. The standout numbers belong to Claude and Gemini, both showing substantial reduction in unnecessary discovery overhead.

Version 3.0.0 Drops with Agent-First Structure

The spec just hit v3.0.0 with two changes that matter for MCP integration: a [MCP] flag on line one lets agents confirm server availability instantly, and Format B introduces a capabilities index where the root file is roughly 150 tokens while individual capability files are around 350 tokens. That's a total of roughly 500 tokens compared to thousands from a typical tools/list response.

Key Takeaways

  • Blueprint Protocol is an open standard (MIT licensed) requiring no vendor tooling—just add blueprint.txt to your app root
  • For MCP servers specifically, it eliminates the tools/list discovery call entirely by providing tool names and parameters upfront
  • Claude showed the biggest improvement at 78% overhead reduction; Grok's 0% reflects that not all models follow discovery signals yet
  • v3.0.0 introduces Format B with a capabilities index for more granular agent routing to specific capability files

The Bottom Line

This is exactly how open standards should emerge—someone hits a wall, builds the fix, publishes it MIT-licensed, and lets the ecosystem decide. Blueprint Protocol isn't trying to replace MCP; it's filling the gap where agents need to discover your app before they can use your tools. If you're shipping web apps or running an MCP server, ten minutes with blueprint.txt might save you hours of debugging agent failures later.