What's New in v0.7.0

Welcome to the v0.7.0 release of mcp-server-cloud-fs! This release evolves the server into an AI-native cloud interface — giving LLM agents smarter tools, stronger safety guarantees, and fewer wasted API calls.

Here are the major highlights.

---

🎯 Dynamic Tool Surface Reduction

When OAuth scopes are granted to a session, the server now filters the tool list so clients only see tools they're authorized to use.

A read-only client (cloud-fs:read) will never see write_file, delete_file, or shell in tools/list. This:

• Reduces prompt token overhead (fewer tool descriptions)
• Prevents tool hallucination (the LLM can't call what it can't see)
• Enforces least-privilege access at the MCP layer

Backwards-compatible: if no grantedScopes are passed, all tools remain available.

Read more about OAuth Scopes

🛡️ DLP Content Sanitization

A new Data Loss Prevention (DLP) middleware redacts sensitive content from tool responses before they reach the MCP client. Ships with 9 default patterns:

Pattern	Example
AWS Access Key	AKIAIOSFODNN7EXAMPLE → [REDACTED:AWS_KEY]
Email Address	alice@example.com → [REDACTED:EMAIL]
US SSN	123-45-6789 → [REDACTED:SSN]
Credit Card	4111 1111 1111 1111 → [REDACTED:CC]
JWT Token	eyJhbG... → [REDACTED:JWT]
API Key	sk-abc123..., sk_live_... → [REDACTED:API_KEY]

Enable via the --enable-dlp CLI flag. Custom patterns can be added programmatically.

Read more about DLP

🧠 AI-Native Tools

Two new tools that let LLMs understand file structure without downloading entire files:

get_file_schema

Extracts structural metadata server-side:

• CSV: Column names, inferred types (string/number/boolean), sample values, row count
• JSON: Root type, keys, nested shapes
• Text: Line count, byte size

summarize_file

Returns a compact overview of any file:

• First and last 5 lines (head/tail preview)
• File size and line count
• Content type

These tools eliminate the common LLM anti-pattern of reading a 10,000-line CSV just to learn its column names.

View the Tool Reference

🔒 Optimistic Concurrency (ETags)

Every file write now computes a SHA-256 content hash (ETag) that's persisted in the VFS inode overlay.

• read_text_file includes the ETag in response metadata: [etag: abc123...]
• edit_file accepts an optional expected_etag parameter
  • Match → edit proceeds normally
  • Mismatch → returns a conflict error with the current ETag
  • Omitted → backwards-compatible, no check

This enables multi-agent workflows where agents can detect and resolve conflicts instead of silently overwriting each other's work.

Read more about ETags

🩹 patch_file Macro Tool

A new high-level tool that combines read + transform + write into a single tool call, reducing the typical 3-step workflow to 1.

Supports two patch formats:

Format	Syntax	Best for
unified (default)	Standard @@ -1,3 +1,3 @@ hunks	Multi-hunk diffs, familiar to LLMs
line_replace	startLine:endLine followed by replacement text	Simple line-range replacements

Both formats support optional expected_etag for concurrency safety.

# Example: unified diff
patch_file path=s3://bucket/config.json patch="@@ -2,1 +2,1 @@
-  \"port\": 3000
+  \"port\": 8080"

View the Tool Reference

---

For a full list of commits and changes, check out the Changelog.