@agentsid/sdk

Identity and auth for AI agents. Drop-in SDK for MCP servers.

• Agent Identity — unique ID per agent instance, not shared API keys
• Per-Tool Permissions — search_* allowed, delete_* denied. Default deny.
• Delegation Chains — every agent traces back to the human who authorized it
• Audit Trail — every tool call logged with tamper-evident hash chain

Install

npm install @agentsid/sdk

Quick Start

import { AgentsID, createHttpMiddleware } from '@agentsid/sdk';

const aid = new AgentsID({ projectKey: 'aid_proj_...' });

// Register an agent with scoped permissions
const { agent, token } = await aid.registerAgent({
  name: 'research-bot',
  onBehalfOf: 'user_123',
  permissions: ['search_*', 'save_memory'],
});

// Validate tool calls (MCP middleware)
const middleware = createHttpMiddleware({ projectKey: 'aid_proj_...' });
const allowed = await middleware.isAllowed(token, 'save_memory'); // true
const denied = await middleware.isAllowed(token, 'delete_all');   // false

MCP Middleware

const middleware = createHttpMiddleware({
  projectKey: 'aid_proj_...',
  baseUrl: 'https://agentsid.dev',
});

// In your MCP tool handler:
const result = await middleware.validate(bearerToken, 'tool_name', params);
// Throws PermissionDeniedError if blocked

API

| Method | Description | |--------|-------------| | registerAgent(opts) | Register a new agent identity | | getAgent(id) | Get agent details | | listAgents(opts?) | List all agents | | revokeAgent(id) | Revoke agent + all tokens | | setPermissions(id, rules) | Set permission rules | | checkPermission(id, tool) | Check if a tool is allowed | | validateToken(token, tool?) | Validate a token | | getAuditLog(opts?) | Query the audit trail |

Links

• Website: agentsid.dev
• Docs: agentsid.dev/docs
• Dashboard: agentsid.dev/dashboard

License

MIT