Lambda CLI – MCP Server for Lambda GPU Cloud

TL;DR

TL;DR: Lambda CLI is an open-source Rust tool that gives you a terminal interface and an MCP server for launching, listing, and terminating Lambda GPU instances — letting you manage cloud GPUs directly from Claude Code with plain English prompts.

Source and Accuracy Notes

• Project page: lambda.ai
• Source repository: github.com/Strand-AI/lambda-cli
• License: MIT (verified via GitHub API license.spdx_id)
• HN launch thread: news.ycombinator.com/item?id=46621786
• Source last checked: 2026-07-18 (commit main branch)

What Is Lambda CLI?

Lambda CLI is an unofficial Rust-based CLI and MCP server for Lambda, a cloud GPU provider. It exposes Lambda’s GPU instance lifecycle — list, start, stop, find — as both direct terminal commands and as MCP tools that AI assistants like Claude Code can调用.

The CLI (lambda) is a traditional command-line tool. The MCP server (lambda-mcp) follows the Model Context Protocol, letting you add it as a backend to any MCP-compatible AI assistant. Once configured, you can manage GPU infrastructure with prompts like “Launch an H100 instance with my SSH key” instead of memorizing API flags.

This is the key difference from Lambda’s own API: instead of writing scripts or hitting REST endpoints, you describe what you want in natural language and let the LLM translate that into API calls.

Setup Workflow

Prerequisites

• A Lambda account with API key access
• Rust toolchain (for source build) or Homebrew (for macOS/Linux binary)
• Node.js 18+ (for npx MCP mode)

Step 1: Install the CLI

Homebrew (macOS/Linux):

brew install strand-ai/tap/lambda-cli

From source:

cargo install --git https://github.com/Strand-AI/lambda-cli

Pre-built binaries are available on the GitHub Releases page.

Step 2: Get Your API Key

Generate an API key from the Lambda dashboard.

Set it as an environment variable:

export LAMBDA_API_KEY=your_key_here

Or use a secret manager command (works with 1Password, pass, and similar):

export LAMBDA_API_KEY_COMMAND="read op://Personal/Lambda/api-key"

Step 3: Verify the CLI Works

lambda list

This outputs all available GPU instance types with per-region pricing and real-time availability. Example output line:

gpu_1x_h100     H100 SXM5 80GB     $2.89/hr   Available in us-east-1

Step 4: Launch Your First Instance

lambda start --gpu gpu_1x_a10 --ssh my-key

Use lambda find to poll until a specific GPU type is available, then auto-launch:

lambda find --gpu gpu_1x_h100 --ssh my-key

Step 5: Connect to Your Instance

lambda running

This prints SSH connection details for all running instances:

i-abc123  A10G  12GB  running  54.82.x.x  ssh [email protected] -i ~/.ssh/my-key.pem

Terminate when done:

lambda stop i-abc123

MCP Server Setup

The MCP server lets Claude Code (or any MCP-compatible AI assistant) manage your GPU infrastructure directly.

Quick Start with npx

No installation required — run directly via npx:

npx @strand-ai/lambda-mcp

Add to Claude Code

claude mcp add lambda -s user -e LAMBDA_API_KEY=your_key -- npx -y @strand-ai/lambda-mcp

With 1Password:

claude mcp add lambda -s user -e LAMBDA_API_KEY_COMMAND='read op://Personal/Lambda/api-key' -- npx -y @strand-ai/lambda-mcp

Then restart Claude Code.

Available MCP Tools

| Tool | Description | |------|-------------| | list_gpu_types | All available GPU types with pricing, specs, and availability | | start_instance | Launch a new GPU instance | | stop_instance | Terminate a running instance | | list_running_instances | Show all running instances with connection details | | check_availability | Check if a specific GPU type is available right now |

Example Prompts Once Configured

Once the MCP server is connected, try these in Claude Code:

• “What GPUs are currently available on Lambda?”
• “Launch an H100 instance with my SSH key ‘macbook’”
• “Show me my running instances”
• “Check if any A100s are available in us-east-1”
• “Terminate instance i-abc123”

Notifications

Lambda CLI can notify you on Slack, Discord, or Telegram when your instance is ready and SSH-able.

Set one or more webhook environment variables before launching:

export LAMBDA_NOTIFY_SLACK_WEBHOOK="https://hooks.slack.com/services/T00/B00/XXX"
export LAMBDA_NOTIFY_DISCORD_WEBHOOK="https://discord.com/api/webhooks/123/abc"
export LAMBDA_NOTIFY_TELEGRAM_BOT_TOKEN="***"
export LAMBDA_NOTIFY_TELEGRAM_CHAT_ID="123456789"

The MCP server sends these notifications automatically when instances become SSH-able — no extra flags required.

Security Notes

• The CLI is unofficial — it is a community project, not affiliated with Lambda Labs
• API keys are stored only in environment variables or via secret manager command — never hardcoded
• LAMBDA_API_KEY_COMMAND executes at startup; ensure your secret manager has appropriate access controls
• The MCP server defers LAMBDA_API_KEY_COMMAND execution until the first API call by default (use --eager to change this)

FAQ

Q: Is this officially supported by Lambda? A: No. Lambda CLI is a community-built tool. Use the official Lambda API for production workloads.

Q: Does it support all Lambda GPU regions? A: The lambda list command shows per-region availability. Check the output for your target region before launching.

Q: Can I use this with Cursor or other MCP-compatible editors? A: Yes. The MCP server is a standard MCP server. You can install it in any MCP-compatible AI assistant using the same npx @strand-ai/lambda-mcp command.

Q: How is this different from Lambda’s own API? A: Lambda’s API is REST-based — you write code or scripts to call it. Lambda CLI wraps that API in both a human-readable CLI and an MCP server. The MCP layer is the main differentiator: you can manage infrastructure via natural language instead of memorizing endpoint paths and request shapes.

Conclusion

Lambda CLI fills a specific gap: if you use Lambda cloud GPUs and want to manage them from an AI assistant, there was no tool purpose-built for that workflow. The MCP layer is the key feature — it makes GPU provisioning feel like a conversation rather than an API call. Installation is straightforward (Homebrew or npx), and the notification integrations mean you do not have to babysit instance startup.

For developers who spin up GPU instances frequently — for model fine-tuning, inference, or batch workloads — this eliminates the context-switching between dashboard, terminal, and AI session.

If you want to try it, start with brew install strand-ai/tap/lambda-cli and lambda list to see what is available in your preferred region.