Universal AI Agent Gateway SDK. Connect any MCP client to 9000+ tools through a single authenticated endpoint. Identity & security layer for AI agents.
AgentDevX SDK
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
**Give your AI hands.**
The identity and security infrastructure every AI agent needs. Not a tool collection — a gateway. Register any API via OpenAPI spec, and every agent gets cryptographic identity, encrypted credential storage, automatic credential injection, and a full audit trail.
What AgentDevX Is
Every developer building an AI agent hits the same wall: auth spaghetti, credential sprawl, integration hell, no stable agent identity, no persistent memory. AgentDevX solves all of this in one gateway.
| Problem | AgentDevX Solution |
|---------|-------------------|
| Auth spaghetti | One Ed25519 cryptographic identity per agent |
| Credential sprawl | AES-256-GCM encrypted vault, scoped per agent |
| Integration hell | Register any API via OpenAPI spec, call via proxy |
| No agent identity | Stable cryptographic identity across all tools |
| No memory | Encrypted per-agent memory, persists across sessions |
Why It's Different
The only gateway an AI agent can discover, install, and use **entirely on its own** — no human OAuth, no manual key exchange, no dashboard.
| Feature | AgentDevX | Composio/Nango | Auth0 |
|---------|-----------|----------------|-------|
| AI self-bootstrap | ✅ | ❌ | ❌ |
| Ed25519 identity | ✅ | ❌ | ❌ |
| Encrypted vault | ✅ AES-256-GCM | ✅ OAuth-based | ✅ Enterprise |
| Per-agent memory | ✅ | ❌ | ❌ |
| MCP native | ✅ | ❌ | ❌ |
| Indie pricing | ✅ Free tier | ❌ | ❌ |
Quick Start
Bootstrap Your Agent
One call. No human needed.
Returns: JWT + 75,000 free credits + instant access to all registered tools.
Connect via MCP
Works with Claude Desktop, Cursor, Windsurf, and any MCP-compatible client.
What Your Agent Gets
- 🔐 **Ed25519 identity** — cryptographic, not OAuth tokens
- 🔒 **AES-256-GCM vault** — store any API key, auto-injected on every call
- 🛡️ **OPA policy engine** — access control, dry-run mode, capability tags
- 📋 **Immutable audit log** — every tool call recorded
- 🧠 **Persistent memory** — encrypted per-agent state across sessions
- ⚡ **42+ pre-registered tools** — plus register your own via OpenAPI spec
- 🌐 **MCP-native** — auto-discovery, protocol 2025-03-26
API Reference
| Endpoint | Method | Description |
|----------|--------|-------------|
| /agents/v1/bootstrap | POST | Create agent + get JWT + 75k credits |
| /agents/me | GET | Agent profile + credit balance |
| /agents/me/memory | POST/GET | Read/write encrypted memory |