App System
AgentLed's integration framework. 100+ apps (LinkedIn, Hunter, CRMs, AI models, web scraping, HTTP) connected through a unified credential and credit system.
What is the App System?
Every integration in AgentLed is an app. Apps expose one or more actions (e.g., Hunter exposes “Find Email by Domain”, LinkedIn exposes “Get Company Profile”). Workflow steps reference apps and actions by ID. The App System handles authentication, credential storage, and credit accounting for every call.
You never write API keys into workflow code. Credentials are connected once per workspace (or per user for OAuth) and reused automatically across all workflows.
App Categories
Data & Enrichment
Hunter, Specter, Affinity CRM, web scraping, news APIs. Usually API-key authenticated and credit-based per lookup.
Social & Communication
LinkedIn, Gmail, Outlook, Slack, Google Calendar. OAuth 2.0 — each user authenticates their own account; credentials are user-scoped.
AI Models
Claude, GPT-4, Gemini, Mistral, and others. Used in AI steps. Billed per token through your workspace credits.
Automation & HTTP
HTTP Request (call any REST API), webhooks, AgentLed internal actions (KG read/write, memory, workflow control). Most require no credentials.
Authentication Methods
| Auth type | Scope | How to connect |
|---|---|---|
| OAuth 2.0 | Per-user | Click Connect in Integrations → authorize in browser |
| API Key | Workspace | Paste key in Integrations → app → Configure; stored encrypted |
| AgentLed-managed | Workspace | No key needed — AgentLed supplies the API key; usage billed to your credits |
| None | — | Public APIs — no setup required |
Discovering Apps via MCP
Use list_apps to get all available integrations, then get_app_actions for the full action schema of a specific app:
// List all available apps
list_apps()
// Returns: [{ id, name, description, category, connected: boolean }]
// Get all actions for an app
get_app_actions("hunter")
// Returns: [{
// id: "hunter.find-email",
// name: "Find Email by Domain",
// inputs: [...],
// outputs: [...],
// creditCost: 1
// }]Credits
Each action has a creditCost (0–15 credits per call). Credits are deducted from the workspace balance when an action completes. Two billing models:
- •per-usage — charged once per action call, regardless of result (e.g., HTTP requests).
- •per-result — charged only when the action returns a result (e.g., Hunter email finder: no charge if no email found).
Credit deductions are visible per-step in the execution timeline. Apps with creditCost: 0 are free to call (e.g., AgentLed internal actions, no-auth HTTP).
Next Steps
- Custom Actions — Call any REST API not in the catalog
- App Actions Reference — Browse input/output schemas for all actions
- MCP Connection — Access app actions from Claude Code, Cursor, or Codex
