E2B MCP

An MCP server on top of E2B: cloud sandboxes for safe code execution by AI agents — 45 tools, stdio and HTTP.

E2B MCP covers the core of the public E2B API with a single MCP server: sandbox lifecycle, file system (with watching and presigned URLs), commands and background processes, PTY, Code Interpreter with contexts, networking (preview URLs), metrics, pause/resume and snapshots, custom templates. It works over stdio and Streamable HTTP (bearer authorisation with constant-time comparison, healthcheck, Docker), with retries on transient errors and output isolation between sandboxes. The v1.0.0 release has been built locally; deployment on a DEV VPS has been verified.

The task

To give an AI agent a safe, isolated environment for running code without giving it access to the user’s own machine.

What’s inside

• 45 tools: sandboxes, files (+watch, binary reads, presigned URLs), commands, Code Interpreter (+contexts), processes, PTY, networking, metrics, snapshots, templates
• Transports: stdio and Streamable HTTP; bearer token, comparison via hmac.compare_digest, /health
• Retries with exponential back-off on RateLimit/Timeout; output isolation between sandboxes; cleanup of streams and subscriptions
• Docker image; a single import point for the E2B SDK (an adapter layer)

Machine translation — being proofread.