CAP Is Deprecated. Read more at CAP - the idea that never took off.
CAP is a standardized protocol designed to provide AI agents knowledge about the structure and organization of a codebase. With CAP, developers can enable AI to communicate with the codebase more efficiently through custom CAP MCP (Codebase Awareness Protocol’s Model Context Protocol) servers.
Three key principles of CAP are:
- Declarative over discovery - You know your codebase better than anyone else, but AI doesn’t. By letting AI read your code, it often ledas to it creating a lot of missing links within its memory, which then leads to broken code development. Write it down once, in plain YAML, let AI direct itself based on that.
- Protocol over prompts - Structured MCP tool responses beat pasted context every time. AI can use only the tools it needs, which helps with automation and time efficiency.
- Zero lock-in - YAML files live in your repo. No SaaS, no accounts, no telemetry.
While CAP was developed only with good intent, the reality was that it wasn’t adopted. Anthropic’s Skills technology took over as it was easier to use. This alone gave me 3 valuable lessons:
- The AI ecosystem is moving fast, and it’s hard to keep up with the latest developments. Evene harder is to push your own technology in front of the AI giants.
- I have a good understanding of when a certain technology is going to storm and what the world needs. I just don’t have enough resources to market it too well.
- Even though CAP wasn’t used, and is beyond its lifecycle of support, it was a great learning experience for me. I learned a lot about AI, MCP servers and even got to make my own VSCode extension. Every bad thing tends to end on a positive note.
The project is well structured and documented, configuration is intended to be shared, provisions all the file structure needed and is ready to be used. If you want to use it, you can find the project on GitHub.
Read more about CAP and how it was developed here.
