Filesystem

Filesystem

All file and directory operations use Claude Code's built-in tools. No MCP server needed — native tools are faster, more capable, and cost zero context tokens when idle.

Quick Reference

Reading Files

Read a Single File

Use the Read tool with an absolute path:

textCopy code
Read: /path/to/file.py

For large files, use offset and limit to read specific sections:

textCopy code
Read: /path/to/file.py (offset: 100, limit: 50)

This reads 50 lines starting from line 100. Use this for files with thousands of lines to avoid flooding context.

Read Multiple Files in Parallel

Issue multiple Read calls in a single response. Claude Code executes them concurrently:

textCopy code
Read: /path/to/file1.py
Read: /path/to/file2.py
Read: /path/to/file3.py

Parallel reads are faster than the MCP's read_multiple_files which serialized internally.

Read Images and PDFs

The Read tool handles binary formats:

• Images (PNG, JPG, SVG): displayed visually
• PDFs: extracted text; use pages: "1-5" for large documents (max 20 pages per call)

Writing and Editing Files

Write a New File

Use the Write tool to create a file or overwrite an existing one:

textCopy code
Write: /path/to/new-file.py
Content: <full file content>

The Write tool requires reading the file first if it already exists. For new files, write directly.

Edit an Existing File

Use the Edit tool for surgical modifications — replace exact string matches:

textCopy code
Edit: /path/to/file.py
old_string: "def old_function():"
new_string: "def new_function():"

The Edit tool fails if old_string is not unique in the file. Provide enough surrounding context to make the match unique, or use replace_all: true for find-and-replace across the entire file.

Prefer Edit over Write for existing files. Edit preserves everything outside the changed region and shows a clear diff. Write replaces the entire file.

Finding Files

By Name Pattern (Glob)

textCopy code
Glob: **/*.py           → all Python files recursively
Glob: src/**/*.ts       → TypeScript files under src/
Glob: *.md              → Markdown files in current directory
Glob: **/test_*.py      → test files anywhere in the tree

Results are sorted by modification time (most recent first).

By Content (Grep)

textCopy code
Grep: pattern="def process_data" type="py"
Grep: pattern="TODO|FIXME" glob="*.py"
Grep: pattern="class.*Controller" output_mode="content" -C=2

Directory Listing

Simple listing:

bashCopy code
ls -la /path/to/directory

Recursive tree with fd:

bashCopy code
fd . /path/to/directory --type f

Tree with depth limit:

bashCopy code
fd . /path/to/directory --type f --max-depth 2

Filter by extension:

bashCopy code
fd -e py /path/to/directory

File Operations

Create Directory

bashCopy code
mkdir -p /path/to/new/directory

The -p flag creates all intermediate directories. Always verify the parent path exists first with ls.

Move / Rename

bashCopy code
mv /path/to/source.py /path/to/destination.py

Rename a file (same directory):

bashCopy code
mv /path/to/old-name.py /path/to/new-name.py

Move a directory:

bashCopy code
mv /path/to/source-dir /path/to/destination-dir

Copy

bashCopy code
cp /path/to/source.py /path/to/destination.py
cp -r /path/to/source-dir /path/to/destination-dir

Delete

bashCopy code
rm /path/to/file.py
rm -r /path/to/directory

Always confirm with the user before deleting files or directories.

File Metadata

bashCopy code
stat /path/to/file.py
ls -la /path/to/file.py
wc -l /path/to/file.py

Common Workflows

Find and Replace Across Files

bashCopy code
# Find all files containing the old string
Grep: pattern="old_function_name" type="py" output_mode="files_with_matches"

# Then Edit each file
Edit: /path/to/file1.py (old_string → new_string, replace_all: true)
Edit: /path/to/file2.py (old_string → new_string, replace_all: true)

Explore an Unfamiliar Codebase

1. Check project structure: fd . --type f --max-depth 2
2. Read configuration: Read: package.json or Read: pyproject.toml
3. Find entry points: Grep: pattern="def main|if __name__" type="py"
4. Read key files identified above

Find Large Files

bashCopy code
fd --type f --exec stat -f '%z %N' {} \; | sort -rn | head -20

Error Handling

Limitations

• Read returns up to 2000 lines by default. Use offset/limit for larger files.
• Read truncates lines longer than 2000 characters.
• Edit requires exact string matching — whitespace and indentation must match precisely.
• Write overwrites the entire file. No append mode. To append, read first, then write the combined content.
• Glob only matches files, not directories. Use Bash ls or fd to list directories.
• PDF reading is limited to 20 pages per call. Specify page ranges for large documents.

Calibration Rules

1. Read before Edit. Always read a file before editing it. The Edit tool enforces this.
2. Edit over Write for existing files. Edit is surgical and shows diffs. Write is a full replacement — use it only for new files or complete rewrites.
3. Glob over Bash for file search. Glob is optimized for pattern matching. Only fall back to fd or find for queries Glob cannot express (size filters, date filters).
4. Grep over Bash for content search. Grep is optimized for ripgrep-based search with proper permissions. Never use grep or rg via Bash.
5. Parallel reads for multiple files. Issue all Read calls in a single response for concurrent execution.
6. Always use absolute paths. Claude Code tools require absolute paths. Never pass relative paths to Read, Write, or Edit.