KiwiFS gives agents four ways to access markdown, from the most native (filesystem) to the most structured (MCP).
Filesystem access
When your agent has a real filesystem mount (NFS, FUSE, or local):
cat /kiwi/concepts/authentication.md
grep -r "timeout" /kiwi/
ls /kiwi/reports/
echo "# New finding" > /kiwi/reports/finding-042.md
REST API
When a filesystem mount isn’t available:
# Read a file
curl localhost:3333/api/kiwi/file?path=concepts/auth.md
# Write a file (with attribution)
curl -X PUT 'localhost:3333/api/kiwi/file?path=reports/finding.md' \
-H "X-Actor: agent:exec_abc" \
-H "X-Provenance: run:run-249" \
-d "# Finding..."
# Search
curl 'localhost:3333/api/kiwi/search?q=timeout'
MCP (Model Context Protocol)
The most structured interface, designed for Claude, Cursor, and other AI tools:
kiwifs mcp --root ~/knowledge # in-process, no server needed
kiwifs mcp --remote http://host:3333 # proxy to a running server
Access protocols
Every protocol flows through the same storage layer. Every write — regardless of how it enters — gets a git commit, a search index update, and an SSE broadcast.
| Protocol | Use case | Example |
|---|
| REST API | Web frontend, scripts | curl localhost:3333/api/kiwi/file?path=index.md |
| MCP | AI agents (Claude, Cursor) | kiwifs mcp --root ~/knowledge |
| NFS | Docker, Kubernetes (native mount) | docker run --mount type=nfs,... |
| S3 | Backup, data pipelines | aws s3 sync s3://knowledge/ /backup/ |
| WebDAV | Windows mapped drives, legacy tools | Map Network Drive in Explorer |
| FUSE | Developer workstations | kiwifs mount --remote http://server:3333 ~/kiwi |
Provenance tracking
Know which agent run produced which content:
curl -X PUT localhost:3333/api/kiwi/file?path=report.md \
-H "X-Actor: agent:exec_abc" \
-H "X-Provenance: run:run-249" \
-d "# Run 249 Report..."
derived-from into the frontmatter automatically. Query later: “show me every page produced by run-249.” Last modified on May 22, 2026
