Filesystem for AI coding assistants

Pick 3 of 5 for AI coding assistantsOfficialAnthropic (reference)

For AI coding assistants, the reference Filesystem server is our third pick of five, and it covers the most basic gap of all: a model that can write code but cannot touch your files. This Anthropic reference server gives the assistant scoped read, write, edit, and search access to local files within directories you allow, which is the core of working in a codebase.

It sits third because the other picks add reach the assistant also needs. Filesystem is the local-files layer, foundational but narrow on its own.

How Filesystem fits

The tools map directly to editing a project: read_text_file and read_multiple_files to take in code, write_file and edit_file to change it, where edit_file applies selective pattern-matched edits with a git-style diff preview. search_files finds matches across the tree, list_directory, list_directory_with_sizes, and directory_tree map the layout, and move_file, create_directory, and get_file_info round out file operations. read_media_file handles images and audio when a task touches binary assets. Access is bounded to allowed directories, so the assistant cannot wander outside them.

Where the siblings take over: Context7 supplies version-accurate library documentation so the assistant codes against the real API, GitHub adds repository operations like reading remote files and managing pull requests, Sequential Thinking gives a structured reasoning scratchpad for tricky refactors, and Playwright drives a browser to verify the UI. Filesystem does none of those; it only operates on local files. Combine it with Context7 and GitHub for a setup that reads docs, edits the local tree, and ships the result.

Tools you would use

ToolWhat it does
read_text_fileRead the complete contents of a file as text, with optional head or tail line limits.
read_media_fileRead an image or audio file and return its base64-encoded data with a MIME type.
read_multiple_filesRead the contents of multiple files simultaneously.
write_fileCreate a new file or overwrite an existing file with the given contents.
edit_fileMake selective edits to a file using advanced pattern matching, with a git-style diff preview.
create_directoryCreate a new directory or ensure one already exists, including parent directories.
list_directoryList the contents of a directory with [FILE] and [DIR] prefixes.
list_directory_with_sizesList directory contents with [FILE] and [DIR] prefixes including file sizes.
directory_treeGet a recursive JSON tree of a directory's contents.
move_fileMove or rename files and directories.
Full Filesystem setup and config →

FAQ

What can the Filesystem server let a coding assistant do?
Read files with read_text_file and read_multiple_files, change them with write_file and edit_file (which previews a git-style diff), search with search_files, and navigate with directory_tree and list_directory, all within directories you explicitly allow.
Why is Filesystem only the third pick for coding assistants?
Because it covers local files and nothing else. Context7 adds current library docs, GitHub adds repository and pull-request operations, and Playwright verifies the UI. Filesystem is the foundational local layer; the higher picks extend what the assistant can reach.