Every AI coding tool wants project context.
Claude Code has its files. Cursor has rules. Codex reads AGENTS.md. OpenCode and Windsurf have their own conventions too.
Those files help. The problem starts when the same project facts get copied everywhere.
One stack change lands in one file. Another file stays old. One agent learns the project. Another starts from stale context.
That drift is what Agent Project Context tries to reduce.
APC proposes one repo-owned context layer in .apc/.
It can hold project metadata, agent definitions, reusable skills, curated memory and MCP registry hints.
It should not hold private chats, raw sessions, secrets or runtime state.
AGENTS.md is still useful. APC gives larger agent workflows more structure around it.
MCP connects agents to external tools. APC describes durable project context.
Different layers. Useful together.
Docs:
https://agentprojectcontext.com/
Repo:
https://github.com/agentprojectcontext/agentprojectcontext
If you use multiple AI coding tools in the same repo, I would like feedback.
What should be shared context, and what should stay tool-specific?
Top comments (0)