tenki-mcp

A Model Context Protocol server for Tenki Cloud. Give any agent — Claude, Codex, Cursor — a disposable microVM it can create, run code in, read and write files, run git, and expose to the web. Sandboxes boot in ~2 seconds and are billed per second.

Part of making Tenki the execution layer coding agents reach for: the agent writes code, Tenki runs it in isolation, and (with Runners + Code Reviewer) tests and reviews it before it ships.

"Run this Python in a fresh sandbox and tell me what it prints."
        │
        ▼   tenki_run_code
   boots a microVM → runs it → returns stdout → tears it down

Quickstart

npm install
npm run build
export TENKI_API_KEY=tk_your_key_here
node dist/index.js         # speaks MCP over stdio

Use it in Claude Desktop / Cursor

Add to your MCP client config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "tenki": {
      "command": "node",
      "args": ["/absolute/path/to/tenki-mcp/dist/index.js"],
      "env": { "TENKI_API_KEY": "tk_your_key_here" }
    }
  }
}

Once published to npm this becomes "command": "npx", "args": ["-y", "tenki-mcp"].

Related MCP server: Kernel MCP Server

Tools

84 tools — full parity with the Tenki unary API (enforced by a CI parity audit), grouped by domain:

Full per-release breakdown in CHANGELOG.md; the plan through v2.0 is in docs/plans/ROADMAP.md.

Auth

Set one of TENKI_API_KEY or TENKI_AUTH_TOKEN. The header is chosen by token prefix: tk_… → Authorization: Bearer, ory_st_… → X-Session-Token, otherwise a session cookie. Override the endpoint with TENKI_API_ENDPOINT (default https://api.tenki.cloud).

How it works

Tenki's API is ConnectRPC — JSON over HTTP/1.1, not REST. Every control-plane call is POST https://api.tenki.cloud/tenki.sandbox.v1.SandboxService/{Method} with a lowerCamelCase JSON body. Per-session file I/O runs on a separate data-plane endpoint returned at create time, authenticated with a short-lived session certificate. This server owns both transports so the tools stay one-liners.

One sharp edge worth knowing: on the current gateway ExecuteCommand reports status and exit code but does not return output artifacts (the SDK streams output over gRPC, which a plain HTTP client can't speak). So tenki_exec and tenki_run_code capture output by redirecting to files (sh -c '… > out 2> err') and reading them back over the data plane. It's transparent to the caller — you get stdout/stderr inline.

The wire details are ported from the live-verified n8n community node.

Roadmap

Shipped v0.2→v0.7: filesystem completion, session/fleet control, preview URLs, snapshots+volumes, templates+registry, workspace admin. See ROADMAP.md. Still ahead:

• v1.0 — a CI parity-audit that fails the build if any API method lacks a tool; npm publish; MCP-registry listings

• v2.0 — streaming exec + interactive shells + an HTTP/SSE transport (needs a gRPC/Connect-streaming transport, the one thing plain HTTP can't do)

• Binary file transfer via signed artifact URLs; batch file writes; SSH access; snapshot-retention settings

Related

• Tenki Sandbox — the platform: https://tenki.cloud

• n8n-nodes-tenki — Tenki as an n8n node: https://github.com/opencolin/n8n-nodes-tenki

License

MIT