tau

Serverless functions for coding agents — deploy, invoke, and monitor from one CLI. As easy as git.

git for your code. tau for your functions.

Your agent writes a webhook handler, a cron job, a data transform — then hits a wall: AWS consoles, IAM policies, YAML, a deploy pipeline it can't drive. So you become the middleman, clicking through Lambda screens for code you didn't write. TauFunctions is the serverless platform your agent runs itself: one install, and it creates, deploys, invokes, and monitors its own functions — from the terminal, with plain JSON in and out.

Works with: Claude Code · Cursor · Cline · Windsurf · Aider · Codex · any MCP client

Create to invoked function in 60 seconds — watch the full demo.

Install

Terminal

npm install -g taufunctions

The npm package is taufunctions; the command is tau.

Quick Start

bash

# Sign up
tau signup you@example.com --tenant my-project

# Create a function from a file
tau fn create "process-order" -r nodejs20 --file handler.js

# Deploy it
tau deploy <function-id>

# Invoke with a JSON payload
tau invoke <function-id> '{"orderId": "ORD-001"}'

# Watch it run
tau logs <function-id>
tau stats <function-id>

# Full reference
tau --help

No consoles, no YAML, no IAM maze

A function on TauFunctions is exactly what your agent already has in its context: a handler file and a runtime. No zip bundles, no execution roles, no API Gateway wiring, no infrastructure-as-code detour just to run 20 lines of JavaScript.

bash

# The whole deploy story:
tau fn create "resize-image" -r nodejs20 --file resize.js
tau deploy $FN
tau invoke $FN '{"url": "https://example.com/photo.jpg", "width": 800}'

Create → deploy → invoke is three commands, and every one returns structured output your agent can parse with --json.

Iterate like you're editing a local file

Shipping v2 doesn't mean a new pipeline run. Push new code at an existing function and redeploy — the function ID, URL, and history stay put.

bash

tau fn code $FN --file hello.js      # swap in the new handler
tau deploy $FN                       # redeploy
tau invoke $FN '{"version": 2}'      # verify

Monitor everything from the terminal

No CloudWatch spelunking. Invocation counts, durations, error rates, and execution logs are one command each — formatted for humans, --json for agents.

bash

tau stats $FN            # invocations, avg duration, error rate
tau logs $FN --tail 20   # recent executions
tau limits               # your plan's quotas and current usage

Runtimes

bash

tau runtimes

Node.js 18/20/22, Python 3.11/3.12, Go 1.21/1.22, Ruby 3.2, and Rust 1.75. Node.js functions execute today on auto-scaling Docker Swarm containers; the remaining runtimes are rolling out on the same executor.

MCP Server

Prefer tools over a CLI? tau ships an MCP server. Point Claude Code (or any MCP client) at it and your agent gets 12 native tools: create, deploy, invoke, update, delete, logs, stats, runtimes, limits, health — the whole function lifecycle.

Terminal

claude mcp add taufunctions -- tau mcp-serve

For clients that use a JSON config (Cline, Cursor, Windsurf), pass your API key via the TAU_API_KEY environment variable. The MCP server runs outside your project directory, so it will not pick up .tau/config.json:

config.json

{
  "mcpServers": {
    "taufunctions": {
      "command": "tau",
      "args": ["mcp-serve"],
      "env": { "TAU_API_KEY": "tf_your_key_here" }
    }
  }
}

Remote MCP — zero install

No CLI at all? Claude Web, Claude Desktop, Raycast, or any hosted MCP client can connect straight to our remote server. Same 13 tools, same API key, nothing to install:

Code

URL:  https://mcp.taufunctions.com/sse
Auth: Authorization: Bearer YOUR_API_KEY

No key yet? Connect without one — the server boots in onboarding mode with a tau_signup tool that provisions your account and hands your agent a live key in the same session.

Features

Functions — create, deploy, invoke, update, delete; JSON payload from CLI or file
Runtimes — Node.js, Python, Go, Ruby, Rust
Deploy — one command to auto-scaling Docker Swarm, no pipeline
Logs & stats — execution logs, invocation counts, durations, error rates
Limits — plan quotas and live usage from the terminal
MCP server — 13 tools, local (tau mcp-serve) or fully remote (mcp.taufunctions.com): Claude Code, Claude Web, Cursor, Raycast, any MCP client
Agent-first output — every command takes --json for structured, parseable results

Pricing: per-invocation billing, no idle cost. Details.

Per-Project Config

bash

tau login --key YOUR_KEY             # saves to .tau/config.json (project-local)
tau login --key YOUR_KEY --global    # saves to ~/.tau/config.json (global)
tau config                           # show which config is active

Local config overrides global. Add .tau/ to your .gitignore.

Agent Integration

Add to your CLAUDE.md, .cursorrules, .clinerules, .windsurfrules, or AGENTS.md:

Code

## TauFunctions
Use the tau CLI for serverless function management.
Key is in .tau/config.json (auto-loaded).
If not configured: tau login --key YOUR_KEY
Run tau --help for the full command reference.

Why this exists

Every time my agent needed "just a small endpoint," the answer was a detour through cloud consoles it couldn't drive — so I ended up clicking through deploy screens for code I didn't write. TauFunctions is the serverless platform the agent runs itself. It's early and I'm iterating fast: if something's rough or missing, tell me — I read everything.

Documentation

License

Proprietary - Tyga.Cloud Ltd. See LICENSE.