AI integrations: MCP server and agent workspace workflows

[FerdinandSu]

Scope

Track AI-native integration work, especially MCP and workspace export for coding/knowledge agents.

Checklist

• Design and implement a VeryFS MCP server so MCP-compatible AI tools can browse mounted resources, read files, inspect manifests, and materialize workspaces through scoped permissions.
• Provide agent-oriented workspace export profiles for Codex, Claude Code, and other coding or knowledge agents.
• Expose manifest, recent, provenance, cache policy, and version metadata in forms that are efficient for tool calls and safe for large repositories.
• Add examples showing signed user requests, least-privilege mounts, and no-cache or metadata-only remote mounts for privacy-sensitive workflows.

Acceptance Criteria

• MCP server exposes a small, permission-aware tool surface.
• Agent workspace export is documented and reproducible.
• Large repositories can be inspected without naive recursive crawling.
• Privacy-sensitive examples demonstrate signed requests and conservative cache modes.

Source

Roadmap section: docs/roadmap.md / AI Integrations.