The filesystem server is the official reference implementation from the MCP project itself, and the one most people install first. It gives a model read and write access to directories you explicitly allow — nothing outside them — and works out of the box with every major client.
Bring your own
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
No token required
How models use it and what it is built for.
The filesystem MCP server exposes local file operations as tools with a directory allowlist enforced on every call. You pass allowed paths as command-line arguments, or supply them dynamically through Roots, and any operation outside those paths is refused. Inside them, a model can read and write files, create directories, move and rename, search by pattern and read file metadata. Media handling is worth knowing about: image and audio files are streamed back as base64 with the right MIME type, so a model can actually look at a screenshot rather than being handed a byte count. It runs over stdio via npx, with no service to sign up for.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
Read every markdown file in ./docs and list which ones have no front matter.
Find all TODO comments under ./src and group them by file.
Rename these screenshots to match the pattern in the folder README.
Summarise what changed between config.old.json and config.json.
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Filesystem
Claude Sonnet 4.5
Sonnet 4.5 handles multi-file reading and line-based edits accurately, which matters when a wrong edit_file call silently corrupts a source file.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
It is the official reference server from the Model Context Protocol project, published as @modelcontextprotocol/server-filesystem. It gives a model read and write access to directories you explicitly allow, and refuses any operation outside them.
Pass allowed directories as command-line arguments, for example npx -y @modelcontextprotocol/server-filesystem /path/to/project. Clients that support Roots can also supply them dynamically. The list_allowed_directories tool shows exactly what is in scope at any moment.
Yes. Image and audio files are returned as base64 with the matching MIME type rather than as raw bytes, so a model can genuinely inspect a screenshot or listen to a clip instead of just seeing a file size.
The allowlist is a real boundary, but inside it the agent can overwrite files. Point it at a project directory under version control rather than your home folder, so any bad edit is one git diff away from being undone.
It operates on your local disk, so there is nothing to host — the server has to run on the machine holding the files. That is also why it needs no credentials: the security boundary is the directory allowlist, not an API key.
GitHub
Drive GitHub repos, PRs and issues with an AI agent.
Docker
List containers, inspect images and manage your Docker environment with AI.
Playwright
Give AI models real browser control — navigate, click, fill forms and screenshot any page.
Context7
Give your AI agent up-to-date, version-specific library docs.