Skip to content

Claude Desktop Setup

Windows: %APPDATA%\Claude\claude_desktop_config.json macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Linux: ~/.config/Claude/claude_desktop_config.json

{
"mcpServers": {
"filesystem-ultra": {
"command": "C:\\path\\to\\filesystem-ultra.exe",
"args": [
"--compact-mode",
"--max-response-size", "5MB",
"--max-search-results", "50",
"--max-list-items", "100",
"--log-level", "error",
"--cache-size", "200MB",
"--parallel-ops", "8",
"C:\\your\\project\\path\\"
]
}
}
}

ParameterValueDescriptionToken Savings
--compact-mode-Enables minimal responses65-75%
--max-response-size5MBLimits response sizePrevents massive responses
--max-search-results50Limits search results85-95% on searches
--max-list-items100Limits directory listings70-80% on listings
ParameterValueDescription
--cache-size200MBMemory cache size (larger = better performance)
--parallel-ops8Concurrent operations (balance for most systems)
--log-levelerrorOnly log errors (reduces overhead)

The final positional arguments are allowed paths:

  • C:\\your\\project\\path\\ - Your project directory
  • Add multiple paths as needed

Section titled “Profile A: Minimal Token Usage (Recommended)”
{
"args": [
"--compact-mode",
"--max-response-size", "5MB",
"--max-search-results", "50",
"--max-list-items", "100",
"--log-level", "error",
"--cache-size", "200MB",
"--parallel-ops", "8"
]
}

Result:

  • Minimal token usage
  • Faster responses (less serialization overhead)
  • Compact responses
  • Less visual detail

{
"args": [
"--compact-mode",
"--max-response-size", "10MB",
"--max-search-results", "200",
"--max-list-items", "300",
"--log-level", "info",
"--cache-size", "200MB",
"--parallel-ops", "8"
]
}

Result:

  • Good token savings (50-60%)
  • More search results
  • Balance between detail and efficiency

{
"args": [
"--max-response-size", "20MB",
"--max-search-results", "1000",
"--max-list-items", "500",
"--log-level", "info",
"--cache-size", "200MB",
"--parallel-ops", "8"
]
}

Result:

  • Maximum visual detail
  • Complete responses
  • Higher token usage

Use: Profile A (Minimal Token Usage)

When performing many file operations, edits, and searches:

  • 100+ operations per session
  • Frequent searches
  • Large directory listings
  • Priority: Minimize tokens

Use: Profile C (Verbose)

When you need all details:

  • Debugging issues
  • Detailed file analysis
  • Full statistics
  • Priority: Maximum information

Use: Profile B (Balanced)

For typical daily work:

  • Mixed operations
  • Balance between tokens and detail
  • Priority: Balance

Command: “List the contents of C:\project\src”

Without compact-mode:

Directory listing for: C:\project\src
[DIR] components (file://C:\project\src\components) - 0 bytes
[FILE] index.js (file://C:\project\src\index.js) - 1024 bytes
[FILE] app.js (file://C:\project\src\app.js) - 2048 bytes
...

Approximately 350 tokens

With compact-mode:

C:\project\src: components/, index.js(1KB), app.js(2KB), config.json, utils/

Approximately 50 tokens = 85% reduction


Command: “Search for ‘TODO’ in the project”

Without compact-mode:

Found 127 matches for pattern 'TODO':
C:\project\src\file1.js:42
// TODO: implement feature
Context:
| function doWork() {
| // TODO: implement feature
...

8,000+ tokens

With compact-mode:

127 matches (first 20): file1.js:42, file2.js:15, file3.js:88, ... (107 more)

Approximately 150 tokens = 98% reduction


Example 3: Complete Session (100 operations)

Section titled “Example 3: Complete Session (100 operations)”
OperationWithout CompactWith CompactSavings
20x write3,00030090%
30x edit6,00060090%
20x list16,0002,00087%
10x search50,0002,00096%
20x other6,0001,00083%
TOTAL81,0005,90092.7%

After configuring, test these commands in Claude:

Show server_info with action stats

Expected (compact-mode):

ops/s:2016.0 hit:98.9% mem:40.3MB ops:2547

List the contents of [your folder]

Expected (compact-mode):

[path]: file1.txt, file2(5KB), folder/, ...

Write 'test' to C:\temp\test.txt

Expected (compact-mode):

OK: 4B written

Problem: Claude still shows long responses

Section titled “Problem: Claude still shows long responses”

Solution: Verify --compact-mode is in the args without JSON syntax errors.

Solution: Verify the full path in command. Use double backslashes \\ on Windows.

Solution: Add the necessary paths as individual arguments at the end of args.

Problem: Responses too short, missing information

Section titled “Problem: Responses too short, missing information”

Solution: Remove --compact-mode to return to verbose mode with full details.


To enable automatic backups, add the backup directory to allowed paths:

{
"mcpServers": {
"filesystem-ultra": {
"command": "C:\\path\\to\\filesystem-ultra.exe",
"args": [
"--compact-mode",
"--max-response-size", "5MB",
"--backup-dir", "C:\\Backups\\MCP",
"C:\\your\\project\\",
"C:\\Backups\\MCP"
]
}
}
}

Important: The backup directory MUST be in allowed paths.

ParameterDefaultDescription
--backup-dir%TEMP%\mcp-batch-backupsDirectory for backups
--backup-max-age7Days to retain backups
--backup-max-count100Maximum backups to keep
--risk-threshold-medium30.0% change for MEDIUM risk
--risk-threshold-high50.0% change for HIGH risk
--risk-occurrences-medium50Occurrences for MEDIUM risk
--risk-occurrences-high100Occurrences for HIGH risk
{
"args": [
"--compact-mode",
"--backup-dir", "C:\\Backups\\MCP",
"--backup-max-age", "14",
"--backup-max-count", "200",
"--risk-threshold-medium", "40.0",
"--risk-threshold-high", "60.0",
"C:\\your\\project\\",
"C:\\Backups\\MCP"
]
}

Benefits:

  • Automatic protection against code loss
  • Risk validation before massive changes
  • Quick recovery with backup (action: “restore”)
  • Complete audit trail of changes

See Backups and Recovery Guide for more information.


For Claude Desktop with high operation volume:

--compact-mode
--max-search-results 50
--max-list-items 100
--log-level error

This reduces token usage by 65-75% without losing essential functionality.


Last updated: March 2026 Version: 4.0.2