mcp-shell
Provides a secure, token-authenticated command execution tool over MCP Streamable HTTP, with a default-deny allowlist and shell metacharacter rejection.
README
mcp-shell
A native Model Context Protocol server that
exposes a single execute_command tool over the Streamable HTTP transport.
The command runs directly in a subprocess of this server — there is no separate
gateway process and no stdio child.
⚠️ Security first
execute_command is remote command execution by design. This server ships
locked down by default with two independent, mandatory app-layer controls,
which are meant to sit on top of (not replace) OS- and network-level hardening —
defense in depth:
Layer 1 — Bearer token (required).
- Set
MCP_SHELL_TOKEN. The server refuses to start without it. - Every HTTP request must send
Authorization: Bearer <token>; anything else gets 401, checked before any MCP session handling.
Layer 2 — Default-deny command allowlist.
MCP_SHELL_ALLOWED_COMMANDSis empty by default, which means zero commands run out of the box.- A command executes only if its program (the first token, matched by name or basename) is explicitly allowlisted.
- Commands containing shell control characters (
;,&,|,`,$,(),<>, …) are rejected, so an allowlisted program can't be used to chain into a non-allowlisted one.
Operational hardening (your responsibility, and just as important).
- Run the server as a dedicated least-privileged user, never root.
- Keep the default bind of
127.0.0.1and put access control in front of it (an authenticated reverse proxy / tunnel). Do not expose it on a public interface. - Treat the token as a secret; rotate it; scope the allowlist as narrowly as the use case allows.
The tool
| Tool | Arguments | Returns |
|---|---|---|
execute_command |
command (string, required) · timeout_ms (int, optional; default 30000, max 300000) |
stdout, stderr, exit code as text. isError: true on non-zero exit or when the command is rejected by the allowlist. |
Output is capped at 100 KB per stream (truncated with a marker); a command that
exceeds its timeout is killed with a [PROCESS KILLED] note.
Quickstart
npm install
# Token is mandatory; allowlist is default-deny — enable only what you need.
MCP_SHELL_TOKEN="$(openssl rand -hex 32)" \
MCP_SHELL_ALLOWED_COMMANDS="echo,ls,git" \
npm start
Without a token the server exits immediately:
npm start
# [mcp-shell] FATAL: MCP_SHELL_TOKEN is not set. Refusing to start ...
Unauthenticated requests are refused:
curl -s -o /dev/null -w '%{http_code}\n' -X POST localhost:3000/mcp \
-H 'Content-Type: application/json' \
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}'
# 401
MCP clients must send the bearer token on every request (e.g. via the client
transport's request headers: Authorization: Bearer <MCP_SHELL_TOKEN>).
Configuration (environment)
| Variable | Default | Purpose |
|---|---|---|
MCP_SHELL_TOKEN |
— (required) | Bearer token; server refuses to start if unset. |
MCP_SHELL_ALLOWED_COMMANDS |
empty (deny all) | Comma-separated allowlisted program names. |
PORT |
3000 |
Port to listen on. |
HOST |
127.0.0.1 |
Bind address. Keep it loopback unless something else gates access. |
MCP_ENDPOINT |
/mcp |
MCP Streamable HTTP mount path. |
ROUTE_PREFIX |
derived | Prefix for /healthz + /ready (defaults to MCP_ENDPOINT minus a trailing /mcp). |
MCP_SHELL_BIN |
/bin/bash |
Shell used to run commands. |
Endpoints
POST/GET/DELETE ${MCP_ENDPOINT}— MCP Streamable HTTP transport (per-session).GET ${ROUTE_PREFIX}/healthz— liveness.GET ${ROUTE_PREFIX}/ready— readiness: spawns a throwaway shell and confirms a trivial command round-trips.
All endpoints require the bearer token (CORS preflight OPTIONS excepted).
Testing
npm test # node --test
Unit tests cover the helpers and the default-deny allowlist (including the shell-metacharacter bypass guard). Integration tests boot the real server and, over the MCP Streamable HTTP client, prove: an allowlisted command runs, a non-allowlisted command is rejected without executing, and a request with no / invalid token gets 401.
License
Apache-2.0
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
Neon Database
MCP server for interacting with Neon Management API and databases
E2B
Using MCP to run code via e2b.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.