Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Descripción del creador (original en inglés)
A fast, configurable filesystem server for the Model Context Protocol (MCP), written in Rust.
Estos tres nunca se combinan en una sola puntuación: las opiniones, las mediciones y la popularidad son cosas distintas.
Todavía no escribimos un resumen sencillo para este. Abajo están los datos que leímos del registro.
Elige la app de IA que usas y copia la configuración.
Necesitas
¿Qué app usas?
Elige tu app arriba. La recordamos, así que la próxima vez el botón dirá “Instalar en …”.
Abre Claude Desktop → Settings → Developer → Edit Config y pega esto en claude_desktop_config.json
{
"mcpServers": {
"fs-rs": {
"command": "npx",
"args": [
"-y",
"fs-mcp-rs"
]
}
}
}Ejecuta esto en tu terminal
claude mcp add fs-rs -- npx -y fs-mcp-rsPégalo en .cursor/mcp.json de tu proyecto, o en ~/.cursor/mcp.json
{
"mcpServers": {
"fs-rs": {
"command": "npx",
"args": [
"-y",
"fs-mcp-rs"
]
}
}
}Tu navegador te pide permiso para abrir la app; luego la app muestra la configuración y te pide confirmar. Los valores de las claves quedan vacíos para que los completes tú.
Pégalo en .vscode/mcp.json de tu proyecto
{
"servers": {
"fs-rs": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"fs-mcp-rs"
]
}
}
}Tu navegador te pide permiso para abrir la app; luego la app muestra la configuración y te pide confirmar. Los valores de las claves quedan vacíos para que los completes tú.
Este se ejecuta en tu computadora, y ChatGPT solo puede conectarse a servidores en internet. Elige otra app.
Tu elección se guarda en una cookie en este dispositivo; no necesitas cuenta.
Esta configuración se genera a partir de la ficha del creador en el registro. Si el README dice otra cosa, sigue el README.
Se ejecuta en tu computadora; la lista de herramientas aparece en tu app de IA después de instalarlo.
Revisamos información que cualquiera puede ver; no hemos leído el código. Las revisiones que no aplican no se cuentan. · Revisado el 7 nov 2026
El nombre en el registro se verificó con la cuenta de GitHub nihmadev, dueña del repo
Todavía no hemos leído este repo en GitHub
Todavía no hemos leído este repo en GitHub
Este se ejecuta en tu computadora; no hay endpoint en internet
Se ejecuta en tu computadora y usa ajustes locales en su lugar
Se ejecuta en tu computadora; solo podemos leer listas de herramientas de endpoints en internet
Comparado con todos los servidores activos de esta categoría en el registro. Las estrellas solo cuentan cuando quien lo publica es dueño del repo.
Superó 1 de 1 revisiones básicas aplicables. No hemos leído el código. Antes de agregar una clave o un token, revisa qué acceso pide el servidor y usa una clave que tenga solo los permisos que necesitas.
El registro no indica ninguna clave obligatoria, aunque algunos servidores te piden iniciar sesión la primera vez que los usas.
Se ejecuta en tu computadora. Tu app de IA lo inicia cuando hace falta, y los datos no pasan por los servidores del creador (a menos que el propio servidor llame a un servicio en línea).
El repo de GitHub cambió por última vez –, y la última versión (1.2.4) llegó al registro hace 4 meses.
Las apps compatibles con MCP, como Claude Desktop, Claude Code, Cursor y VS Code. También necesitas Node.js en tu computadora.
En la misma categoría: Spec Driven Development, Remote Desktop Commander, SwarmMemo
Servidores de la misma categoría hechos por otros creadores, ordenados por estrellas.
Cómo leer una tarjeta:★ Estrellas de GitHubiTres pilares: reseñas · revisiones · usoiActualizado = fecha en que se publicó la última versión en el registroi
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Talk with agent swarms: rooms, private messages and paid tasks for AI agents
Copiado tal cual del repo del creador. PopNx MCP no lo ha revisado ni lo respalda. Se quitaron HTML, imágenes y scripts. Los enlaces llevan fuera de este sitio.
A fast, configurable filesystem server for the Model Context Protocol (MCP), written in Rust.
fs-mcp-rs gives an MCP client a small set of filesystem tools while keeping access inside directories selected by the operator. It has no implicit filesystem root, no built-in personal path, and no writable default. The server starts only when you provide a configuration file.
The public distribution is one crate and one executable, both named fs-mcp-rs. The protocol, policy, filesystem, search, and settings components remain separate Rust modules inside that crate. This keeps installation simple:
cargo install fs-mcp-rs
Users do not need to install or understand a graph of internal crates. The benchmark runner remains a development-only workspace member and is not included in the crates.io package.
npx fs-mcp-rs..gitignore support and configurable hidden-file handling.[OK] / [WARN]).npx execution, or Rust 1.85+ when building from source.The easiest way to run fs-mcp-rs with any MCP client (such as Claude Desktop or Cursor) is using npx:
npx fs-mcp-rs /path/to/project
The npm package automatically downloads the appropriate precompiled binary for your operating system and architecture (Windows x64, Linux x64, macOS x64/arm64).
cargo install fs-mcp-rs
fs-mcp-rs --version
git clone https://github.com/nihmadev/fs-mcp-rs.git
cd fs-mcp-rs cargo build --release
The executable will be at target/release/fs-mcp-rs on Unix-like systems or target\release\fs-mcp-rs.exe on Windows.
Copy configs/example.toml somewhere appropriate and edit it. The server intentionally has no default config path:
cp configs/example.toml fs-mcp-rs.toml
Windows PowerShell:
Copy-Item configs\example.toml fs-mcp-rs.toml
Minimal read-only configuration:
[server]
host = "127.0.0.1" port = 8000 max_concurrency = 32 max_io_concurrency = 16
[filesystem] roots = ["/home/alice/projects"] read_only = true max_read_bytes = 8388608 max_write_bytes = 8388608 follow_links = false
[search] max_results = 1000 max_concurrency = 4 worker_threads = 4 regex_cache_capacity = 64 include_hidden = false respect_gitignore = true
[terminal] enabled = true max_concurrency = 2 default_timeout_ms = 30000 max_timeout_ms = 300000 max_output_bytes = 4194304
Windows roots use escaped backslashes or forward slashes:
roots = ["C:\\Users\\Alice\\Projects", "D:/Shared/Documentation"]
Relative roots are resolved relative to the directory containing the configuration file, not relative to an arbitrary process working directory.
[server]host: IP address to bind. Keep 127.0.0.1 unless remote clients genuinely need access.port: TCP port for /health and /mcp.max_concurrency: maximum number of MCP requests processed at once.max_io_concurrency: maximum number of simultaneous non-search filesystem jobs.log_tools: whether concise tool execution logging ([OK] / [WARN]) is printed (default: true).[filesystem]roots: non-empty list of allowed directories. Every requested path must resolve inside one of them.read_only: when true, all mutating tools are rejected.max_write_bytes: largest permitted write_file or resulting edit_text content.follow_links: whether paths containing symbolic links may be followed. Leave this false unless you have reviewed the consequences.[search]max_results: maximum matches returned by one search.max_concurrency: maximum simultaneous search jobs.worker_threads: filesystem traversal threads used by each search.regex_cache_capacity: number of compiled search patterns cached in memory; 0 disables the cache.include_hidden: include hidden files and directories.respect_gitignore: apply Git ignore files while walking directories.[terminal]enabled: expose command execution. When disabled, run_command returns a tool error.max_concurrency: maximum terminal commands running simultaneously.default_timeout_ms: timeout used when a call does not provide one.max_timeout_ms: largest timeout a call may request.max_output_bytes: maximum combined stdout/stderr bytes retained per terminal session. Old output is evicted and reported as dropped/truncated.max_read_bytes: maximum output bytes returned by one terminal_read call.max_wait_ms: maximum long-poll duration for one terminal_read call.session_retention_ms: how long completed sessions and their output remain readable before automatic cleanup.run_command executes arbitrary commands through cmd.exe on Windows and /bin/sh on GNU/Linux and macOS. It is advertised with destructive and open-world MCP annotations so compatible providers can show a warning. Enabling it gives commands all permissions of the operating-system account running the server; filesystem roots and read-only mode do not sandbox terminal commands.
Unknown fields and zero-valued safety limits are rejected. This catches misspelled settings instead of silently ignoring them.
Start directly by specifying allowed filesystem directories on the command line:
fs-mcp-rs /home/alice/project1 /home/alice/project2
If no configuration file or positional roots are passed, fs-mcp-rs uses default settings scoped to the current working directory.
Pass the configuration explicitly:
fs-mcp-rs --config /absolute/path/to/fs-mcp-rs.toml
Or use the environment variable:
FS_MCP_CONFIG=/absolute/path/to/fs-mcp-rs.toml fs-mcp-rs
PowerShell:
$env:FS_MCP_CONFIG = "C:\config\fs-mcp-rs.toml"
fs-mcp-rs
--stdio. Structured logs are written to stderr to keep stdout clean for line-delimited JSON-RPC messages./health and /mcp endpoints.Check HTTP health:
curl http://127.0.0.1:8000/health
Expected response:
ok
The MCP endpoint is:
http://127.0.0.1:8000/mcp
The run subcommand always starts the MCP server in HTTP mode and can manage an ngrok, Cloudflare Quick Tunnel, or zrok public share in the same process lifecycle:
fs-mcp-rs run --ngrok /path/to/allowed/directory
fs-mcp-rs run --ngrok --config ./fs-mcp-rs.toml fs-mcp-rs run --cloudflared /path/to/allowed/directory fs-mcp-rs run --zrok /path/to/allowed/directory
The providers are mutually exclusive. On Windows, the tunnel opens in a new console by default. Use --tunnel-window hidden to avoid a separate visible console while keeping the generated URL in the current terminal, or --tunnel-window inherit to attach normally. On Linux and macOS, new inherits the current terminal because creating a terminal window is desktop-specific. Each provider accepts repeatable extra arguments and an explicit executable path:
fs-mcp-rs run --ngrok --tunnel-window hidden --ngrok-arg=--region=eu
fs-mcp-rs run --ngrok --ngrok-bin /opt/ngrok/ngrok fs-mcp-rs run --cloudflared --cloudflared-arg=--no-autoupdate fs-mcp-rs run --zrok --zrok-bin /opt/zrok/zrok
Stopping the MCP server also stops the tunnel process. Existing invocations such as fs-mcp-rs [PATHS], fs-mcp-rs --config FILE, and fs-mcp-rs serve are unchanged. The earlier --ngrok-window spelling remains an alias for --tunnel-window.
For testing with a remote MCP client, keep fs-mcp-rs bound to loopback and expose it through a tunnel. Start the server first:
fs-mcp-rs --config ./fs-mcp-rs.toml
Verify the local endpoint before creating a tunnel:
curl http://127.0.0.1:8000/health
[!CAUTION]
A public tunnel URL gives remote clients a path to the tools and filesystem roots allowed by your configuration. Start with
read_only = true, grant only narrowly scoped roots, never expose secrets, and stop the tunnel when it is no longer needed. Quick tunnels are intended for development—not permanent production hosting.
Install cloudflared, then create a temporary Quick Tunnel:
cloudflared tunnel --url http://127.0.0.1:8000
cloudflared prints a temporary URL similar to https://random-name.trycloudflare.com. Use the MCP endpoint with /mcp appended:
https://random-name.trycloudflare.com/mcp
For a stable hostname, create a named Cloudflare Tunnel, map a DNS hostname to it, and route that hostname to http://127.0.0.1:8000. Protect long-lived deployments with Cloudflare Access or another authentication layer.
Install ngrok, authenticate its CLI if required by your account, and start an HTTP tunnel:
ngrok http 8000
Use the HTTPS forwarding URL shown by ngrok and append /mcp:
https://example.ngrok-free.app/mcp
The local inspection UI is normally available at http://127.0.0.1:4040. It can expose request details, so keep it local and do not share it publicly.
Use the public HTTPS URL as an HTTP/Streamable HTTP MCP server:
{
"mcpServers": { "filesystem": { "transport": "http", "url": "https://your-public-host.example/mcp" } } }
Test both endpoints after the tunnel starts:
curl https://your-public-host.example/health
curl -i https://your-public-host.example/mcp
A non-200 response from a plain `G
Solo se muestra el principio; lee el resto en el repo