E2B to MCP: Manage Cloud Sandboxes, Logs, Metrics, Templates and Snapshots in any AI Agent

💡 No install? Use cloud.anythingmcp.com directly. Sign in, click Connectors → E2B, paste your credentials, mint an MCP API key and you're done. No Docker, no git clone, no local server.

E2B

Manage the E2B cloud sandboxes of your team with your AI: list, start, inspect, pause, resume and extend sandboxes, read their logs and CPU, memory and disk metrics, and browse the team's templates, builds and snapshots.

Setup

Setting up takes about 2 minutes. You need an E2B account; running sandboxes are billed per second to the team (the Hobby plan starts with free credits).

1. Sign in at https://e2b.dev/dashboard and open Keys (https://e2b.dev/dashboard?tab=keys).
2. Copy a team API key. It starts with e2b_; a personal access token (sk_e2b_...) does not work here.
3. Paste it into E2B_API_KEY. The connector sends it in the X-API-Key header, and it acts for the whole team.
4. Install the connector and run e2b_list_sandboxes to check it.

Which tools cost money: e2b_create_sandbox starts a sandbox and e2b_resume_sandbox resumes a paused one. A running sandbox is billed per second for its CPU and memory until it times out, is paused or is killed, and e2b_set_sandbox_timeout keeps it billed for longer when it moves the end time out. The AI asks you before starting, resuming or extending a sandbox, keeps timeout short and pauses sandboxes or lets them expire when the work is done. The read tools are free, and a paused sandbox is not billed for compute.

Templates: e2b_list_default_templates shows E2B's ready-made templates such as base, code-interpreter-v1 and desktop with CPU, memory, disk and envd version; e2b_list_templates lists the team's own templates with aliases, build status and spawn counts; e2b_list_template_builds lists the builds of one template.

Starting a sandbox: e2b_create_sandbox takes template_id (id or alias, default base), timeout in seconds (default 300) and optionally metadata (string key-value pairs to find it again), env_vars, auto_pause (pause at the timeout instead of ending) and allow_internet_access. The answer has the sandboxID and domain. The AI does not put secrets into env_vars that you have not approved.

Inspecting: e2b_list_sandboxes filters by state (running or paused), template, metadata such as user=abc&app=prod and started_after; e2b_get_sandbox shows template, resources, start and end time, state and network settings; e2b_get_sandbox_logs reads system and process logs, filtered by minimum level or a case-sensitive search text; e2b_get_sandbox_metrics shows CPU, memory and disk use over time.

Lifecycle: e2b_set_sandbox_timeout sets a new end time counted from now and replaces the old one, so a shorter value ends the sandbox sooner. e2b_pause_sandbox freezes a sandbox with its files and, by default, its memory (memory: false keeps only the file system, and resuming then reboots it); e2b_resume_sandbox resumes it or extends a running one; e2b_list_snapshots lists the saved snapshots.

Switched off at install: e2b_kill_sandbox. It cannot be undone (processes, memory and files are gone), so a workspace admin turns it on when the team needs it.

Running code is not part of this connector: commands, files and the code interpreter are served by each sandbox itself (its envd agent on https://<port>-<sandboxID>.e2b.app) with its own access token, and the team API key must not be sent into a sandbox where generated code runs. Use the E2B SDKs (e2b, @e2b/code-interpreter) for that, or a tool that wraps them. Sandbox access tokens are removed from the answers.

Data formats: times are ISO 8601, UTC; metrics start and end are Unix seconds and log cursor values Unix milliseconds. cpuCount is vCPUs, memoryMB and diskSizeMB are megabytes.

Paging: e2b_list_sandboxes, e2b_list_templates, e2b_list_template_builds and e2b_list_snapshots take limit (default 50, at most 100) and return the next page token in _headers.x-next-token, with the list itself under data; the AI passes it as next_token until it is missing. e2b_list_sandboxes also returns _headers.x-total-running. e2b_get_sandbox_logs returns up to 1,000 lines (default 100) from a cursor, forward or backward.

Limits: the team has a limit of concurrent sandboxes. timeout is capped by the plan: 1 hour on Hobby, 24 hours on Pro.

Common errors: 400 means a bad parameter (for example a timeout over the plan's maximum); 401 a wrong key, or an access token instead of a team API key; 403 the team may not use the template; 404 an unknown sandbox or template (a sandbox that timed out is gone); 409 the sandbox is already paused or being paused; 429 too many concurrent sandboxes or requests (the AI waits instead of retrying in a loop); 500 and 503 are E2B problems.

The connector does not run code, commands or file operations inside sandboxes and does not build templates.

Prerequisites: E2B_API_KEY

Install the connector locally

curl -fsSL https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml -o docker-compose.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Open http://localhost:3000/connectors/store, pick E2B and paste in the env vars listed under Prerequisites.

Available tools

Example prompts

• "Which E2B sandboxes are running right now, which template does each use, and when does each one end?"
• "Start a code-interpreter-v1 sandbox for 10 minutes with the metadata user=demo and give me its sandbox id."
• "Show me the error logs of sandbox ixk2b9z0."
• "How much CPU and memory has sandbox ixk2b9z0 used since it started?"
• "Pause every running sandbox with the metadata app=staging."

Next steps

• Connect to Claude
• Connect to ChatGPT