Agent-File-System brings progressive disclosure and successful research to MCP servers through a familiar file system metaphor.
Rather than overwhelming AI agents with comprehensive capability
documentation upfront, it provides just two core tools
(read and write) with powerful path
syntax extensions that enable gradual discovery of features,
commands, and domain expertise.
This approach leverages AI agents' existing understanding of file
system navigation while introducing novel capabilities like section
addressing within files, command execution via !
syntax, and semantic code search—all discoverable on-demand through
simple exploration patterns.
MCP endpoint: /mcp
Health check: /health_check