The Basic Flow
Getting a working system takes four steps:1. Create Account & Sign In
Sign up for Noorle and log into the Console (Web UI). What you get:- A personal/organization account
- Access to Web UI for managing everything
- REST API credentials for automation
- Secure storage for your agents, gateways, and configurations
2. Create a Gateway
MCP Gateway (for connecting AI agents)- Gets a unique URL:
mcp-{id}.noorle.com - Exposes tools via MCP protocol
- Perfect for: Claude, custom AI agents, and any MCP-compatible client
- Runs continuously or on-demand
- Can invoke tools, remember context, collaborate with other agents
- Gets a unique URL:
agent-{id}.noorle.com - Perfect for: Customer support, data processing, automation
- Just giving tools to an external agent? → MCP Gateway
- Need the AI to be persistent and autonomous? → Agent
3. Attach Capabilities
Capabilities are tools that your gateway/agent can use. Three types:Built-in (Already Available)
These are pre-built and ready to use:- Files - Read/write files in a sandboxed workspace
- Web Search - Search the internet
- HTTP Client - Make API calls
- Code Runner - Execute Python, Node, or Rust code
- Knowledge Retrieval - Semantic search in your knowledge bases
- Sandbox - Isolated Linux environment
- Browser - Automated web interaction
- Computer - Direct system access (agents only)
Plugins (You Create)
Custom tools written in WebAssembly:- Upload a
.npackfile (your compiled plugin) - Define what tools it exposes
- Deploy new versions without downtime
- Perfect for proprietary logic, security, performance
Connectors (You Link)
Connection to external services:- REST API - Describe endpoints, Noorle learns how to use them (OpenAPI import supported)
- MCP Registry - Use servers from the MCP registry (e.g., Airtable, GitHub)
- Custom MCP - Connect to your own MCP server
4. Connect a Client
Your gateway is now live. Connect clients:For MCP Gateway
Use any MCP client library:For Agent
The agent runs autonomously. You can:- Chat with it via REST API
- Integrate it into your app via webhooks
- Deploy it on messaging platforms (Slack, Telegram, SMS)
Key Concepts Snapshot
| Concept | What It Is | Why It Matters |
|---|---|---|
| Capability | A tool or service (builtin, plugin, or connector) | Defines what your agent/gateway can do |
| MCP Gateway | Virtual endpoint exposing tools via MCP | Standard way to give AI agents tools |
| Agent | Autonomous AI system with memory and tool access | Can act independently, remember context |
| Memory | Three-tier system (working → summary → archive) | Agents remember important info efficiently |
| Omni Tool | AI-powered tool discovery | Agent says “upload this PDF” and omni_tool figures out which tool to use |
| Connector | Link to external service | Reuse existing APIs without code |
| Plugin | Custom tool in WASM sandbox | Run proprietary logic securely |
What Happens When Your Agent Uses a Tool?
Memory: How Agents Remember
Agents have a three-tier memory system designed for efficiency and long-term context:-
Working Memory (Recent)
- Last 10-20 messages
- Fast access, limited capacity
- Used for immediate context in active conversations
-
Message Journal (Permanent)
- Every message preserved
- Immutable history
- Used for compliance and full conversation auditing
-
Summary Memory (Compressed)
- LLM-generated summaries of older conversations
- Oldest messages summarized and archived
- Used to maintain context in long-running conversations
Authentication
Three methods depending on your use case: Sessions (Web UI, browser)- Sign in with email/password
- Cookie-based
- Perfect for: Console, web apps
- Format:
ak-{access_id}{secret} - Long-lived, scoped to account or specific resources
- Perfect for: Automations, integrations
- Device authorization flow for CLI
- Short-lived access + refresh tokens
- Perfect for: CLI tools, mobile apps
- Connect your Noorle account to external services
- Supports dynamic client registration
- Perfect for: SaaS integrations
Multi-Agent Workflows
Agents can collaborate: Agents communicate via A2A protocol. Coordinator routes tasks, agents report back.Deployment Options
| Option | Best For | Effort |
|---|---|---|
| Console | Manual management, testing | Minimal |
| REST API | Programmatic creation, automation | Low |
| Infrastructure as Code | Large deployments, team collaboration | Medium |
| Embedded Agent | Apps that need AI capabilities | High |
Next Steps
Learn Core Concepts
Deep dive into MCP, gateways, capabilities, agents, memory, and more.
5-Minute Getting Started
Step-by-step guide to create your first gateway and agent.
Build a Plugin
Create a custom tool in WebAssembly.
API Reference
Detailed endpoint and type documentation.
Once you understand the flow, explore the Learn tab for deeper knowledge, or jump to the Use tab to start building.