Agent Auth SDK

Overview

AI agents are autonomous — but someone needs to be accountable. The Agent Auth SDK gives humans control over what agents can spend, which tools they can access, and when to pull the plug.

One integration. Every framework. Every payment rail.

ANY FRAMEWORK          →  50c Agent Auth   →  ANY PAYMENT RAIL
(OpenClaw, CrewAI,        (authorization,      (x402, Visa TAP,
 Claude, AutoGPT,          limits, audit,       Mastercard AgentPay,
 OpenAI, Google ADK)       kill switch)         or 50c credits)

Why This Exists

Protocol	What It Does	What It’s Missing
MCP	Connects agents to tools	No payments, no spending limits
A2A	Agents discover each other	No financial layer
x402	Agents pay per HTTP call	No budgets, no human approval

Agent Auth is the missing authorization layer between agents and money.

Quick Start

from fiftyc import AgentAuth
import os

agent = AgentAuth(api_key=os.getenv("FIFTYC_API_KEY"))

@agent.authorize(credit_limit=100, approval_above=50)
def my_agent_task():
    result = agent.call("web_search", query="latest AI papers")
    return result

Core Features

1. Per-Tool Spending Limits

agent.set_limits(
    web_search=50,      # $50/month cap
    genius=200,         # $200/month cap
    titan=0             # Blocked entirely
)

2. Human-in-the-Loop Approval

agent.configure(
    auto_approve_below=10,     # <$10 → agent decides
    require_approval_above=50, # >$50 → human must approve
    kill_after=500             # Hard stop at $500 total
)

3. Immutable Audit Trail

Every action logged with tamper-proof hashes. Export for compliance.

trail = agent.audit_trail(last=100)
trail.export("csv")  # For CISO/compliance

4. Kill Switch

Instantly revoke all agent permissions.

agent.revoke()  # Stops all agent actions immediately

5. Platform-Agnostic Identity

One agent ID works across every framework.

agent = AgentAuth(id="agent_abc123")
# Works in OpenClaw, CrewAI, AutoGPT, raw Python

Integrations

Platform	Integration Type	Status
MCP (Claude, Cursor, OpenClaw)	MCP Server	🔜 Soon
OpenClaw	Skill plugin	🔜 Soon
CrewAI	Python decorator	🔜 Soon
AutoGPT	Python decorator	🔜 Soon
A2A	Agent Card at `.well-known/agent.json`	🔜 Soon
x402	Payment backend	🔜 Soon

MCP Server Integration

{
  "mcpServers": {
    "50c-agent-auth": {
      "command": "npx",
      "args": ["50c-agent-auth"],
      "env": {
        "FIFTYC_API_KEY": "your_key_here"
      }
    }
  }
}

Payment Rails

Agent Auth works with any payment method:

Rail	How	Status
50c Credits	Direct debit from 50c balance	🔜 Soon
x402 (USDC)	HTTP 402 payment protocol	🔜 Soon
Visa TAP	Trusted Agent Protocol	Planned
Mastercard Agent Pay	Agent commerce framework	Planned

Pricing

Agent Auth will be included with all 50c plans. No additional cost for authorization, audit trails, or kill switch functionality.

Feature	Free	Starter	Pro	Enterprise
Agent wallets	1	5	25	Unlimited
Audit trail retention	7 days	30 days	90 days	1 year
Kill switch	✅	✅	✅	✅
Anomaly detection	—	—	✅	✅
Custom approval rules	—	—	✅	✅

The Trust Problem

Only 16% of consumers trust AI to make payments. — Nevermined, 2026

Agent Auth solves trust by keeping humans in control:

Agents operate within human-set boundaries
Every action is auditable and tamper-proof
Humans can kill any agent instantly
Spending limits are enforced server-side (not bypassable)