Capsule Bash Server
OfficialThe Capsule Bash MCP server lets you run bash commands in secure, persistent, isolated WebAssembly sandboxes with no access to the host filesystem or network.
run: Execute a bash command in a sandboxed environment, returning stdout, stderr, exit code, a filesystem diff of what changed, and current shell state (cwd + env vars). Commands sharing the samesession_idmaintain persistent state across calls.reset: Wipe a session's filesystem and shell state (cwd, env vars) back to initial values.sessions: List all currently active shell sessions.
Each session runs in its own WebAssembly address space, ensuring full isolation between sessions with no cross-session data leakage.
Capsule Bash MCP
An MCP server that gives your AI agent the ability to run bash commands in a secure, persistent, sandboxed environment.
How It Works
Each session runs inside a WebAssembly sandbox. The sandbox provides:
Persistent state: cwd, env vars, and filesystem changes persist across commands within a session
Filesystem diff: every
runresponse includes a diff of what changed on diskIsolated memory: each session gets its own address space, no cross-session leakage
No host access: the sandbox cannot reach your host filesystem or network
Learn more about Capsule Bash.
Related MCP server: execkit-mcp
Tools
Tool | Description |
| Run a bash command in a sandboxed session. Returns stdout, stderr, exit code, filesystem diff, and current state (cwd + env). |
| Reset the filesystem and state (cwd, env vars) for a session back to their initial values. |
| List all active sessions. |
Sessions
Commands within the same session_id share cwd, environment variables, and filesystem state across calls.
Example
Ask your AI agent:
"Write a Python script that averages a list of numbers."
The agent calls run sequentially:
{ "command": "mkdir -p /data && cd /data", "session_id": "custom_session" }
{ "command": "echo 'nums = [x for x in [1, 2, 3, []] if isinstance(x, int)]\nprint(sum(nums) / len(nums))' > avg.py", "session_id": "custom_session" }
{ "command": "python3 avg.py", "session_id": "custom_session" }Each call returns stdout, stderr, exitCode, a filesystem diff, and the updated state to enrich context and keep trace in conversation history.
Setup
Add to your MCP client configuration (e.g. Claude Desktop, Cursor):
{
"mcpServers": {
"bash": {
"command": "npx",
"args": ["-y", "@capsule-run/bash-mcp"]
}
}
}Limitations
Not all bash commands and options are implemented. Feel free to open an issue if a command is missing or behaves unexpectedly.
Available Tools
3 toolsresetA
Reset a session's filesystem and shell state (cwd, env vars) to their initial values. Useful to start fresh without creating a new session.
Inputs: session_id.
runB
Execute a Bash command in the sandboxed environment
Inputs: command (required), session_id.
sessionsA
List all active session IDs and their current state.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool has a unique, clear purpose: reset reverts state, run executes commands, and sessions lists active sessions. No ambiguity or overlap exists.
All tool names are single, imperative verbs (reset, run, sessions) following a consistent, simple pattern.
Three tools is well-scoped for a bash sandbox server, covering essential operations without superfluous or missing tools.
The set covers core actions (run, reset, list sessions) but lacks explicit session creation; likely sessions are created implicitly, leaving a minor gap.
Maintenance
Related MCP Connectors
Secure Linux sandboxes for AI agents, as MCP tools.
Cloud Linux VMs for your agent: run code, Docker, files and HTTPS previews in a Sandboxer microVM.
Linux machines for AI agents: create, run commands, snapshot, fork, sandbox. Alpha.
Autonomous agent guardrails, bash sandboxing, MCP rugpull diffing & prompt firewalls.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceA throwaway Docker sandbox for agents to run code and shell commands safely.8 npmMIT
- AlicenseAqualityBmaintenanceStateful, structured, safe shell sessions for AI agents, on real infrastructure.713Apache 2.0
- AlicenseAqualityBmaintenanceExposes persistent, stateful remote Bash sessions to AI agents via SSH, enabling command execution with preserved working directory and environment.772 npm1MIT
- AlicenseAqualityCmaintenanceSandboxed bash execution MCP server for AI agents, using an in-memory virtual filesystem overlay to prevent real filesystem damage, with configurable network access, timeouts, and optional Python/JS runtimes.984 npmApache 2.0