Agentic API
Benchmarked against: Anthropic — (Industry concept alignment) Research: UB entry ub-7b3e018d273c Scope: How SuperPortia's API framework aligns with the Agentic API industry concept
Agentic API is a real industry concept — action-oriented APIs designed for AI agents and autonomous systems. SuperPortia's docs site simultaneously serves as a check on whether our API framework meets this standard.
What is an Agentic API?
An Agentic API is an API designed to be consumed by AI agents, not just human developers. The key differences from traditional APIs:
| Feature | Traditional API | Agentic API |
|---|---|---|
| Consumer | Human developers | AI agents |
| Discovery | Documentation + Swagger | Tool descriptions + schema |
| Invocation | Code integration | Natural language → tool call |
| Error handling | Status codes + messages | Self-correcting retry with reasoning |
| Authentication | API keys, OAuth | Agent identity + ship credentials |
| Orchestration | Application logic | Agent reasoning + planning |
SuperPortia's Agentic API surface
SuperPortia exposes its capabilities through MCP (Model Context Protocol) servers, which are inherently agentic — designed for AI agents to discover and use.
Knowledge management
| Tool | Agentic behavior | Traditional equivalent |
|---|---|---|
search_brain | Agent reasons about what to search | GET /api/search?q= |
ingest_fragment | Agent decides what to ingest | POST /api/entries |
get_entry | Agent retrieves specific knowledge | GET /api/entries/:id |
update_entry | Agent updates knowledge | PATCH /api/entries/:id |
Work orchestration
| Tool | Agentic behavior | Traditional equivalent |
|---|---|---|
create_work_order | Agent creates structured tasks | POST /api/work-orders |
dispatch_work_order | Agent selects engine and delegates | POST /api/dispatch |
complete_work_order | Agent reports results | PATCH /api/work-orders/:id/complete |
factory_floor_status | Agent monitors team | GET /api/status/floor |
Communication
| Tool | Agentic behavior | Traditional equivalent |
|---|---|---|
send_agent_message | Agent-to-agent messaging | POST /api/messages |
check_agent_mailbox | Agent checks for messages | GET /api/mailbox |
agent_heartbeat | Agent registers presence | POST /api/heartbeat |
Intelligence
| Tool | Agentic behavior | Traditional equivalent |
|---|---|---|
intel_search | Agent researches and auto-ingests | POST /api/intel/search |
search_web | Agent delegates web search | POST /api/search/web |
call_model | Agent delegates to lower-cost engine | POST /api/models/invoke |
run_patrol | Agent patrols configured domains | POST /api/patrol/run |
Agentic API design principles
SuperPortia's API follows these agentic design principles:
1. Self-describing tools
Every tool includes a description and JSON Schema that agents can reason about:
{
"name": "search_brain",
"description": "Search Universal Brain using hybrid search (semantic + keyword).",
"parameters": {
"query": {"type": "string", "description": "Search query"},
"limit": {"type": "number", "default": 10}
}
}
2. Autonomous decision points
Agents choose which tools to use based on reasoning, not hardcoded logic:
3. Identity-aware operations
Every API call carries agent identity:
| Field | Purpose |
|---|---|
SP_SHIP_ID | Which ship is making the call |
SP_AGENT_ID | Which agent on that ship |
| Timestamp | When the call was made |
4. Multi-engine dispatch
Agents can route work to different engines based on task requirements:
| Engine | Cost | When agent should choose it |
|---|---|---|
| Groq | Free | Trivial research tasks |
| Gemini | Cheap | Quality research with citations |
| Claude | Expensive | Code operations and architecture |
Agentic API maturity model
How well does SuperPortia's API surface meet agentic standards?
| Level | Description | SuperPortia status |
|---|---|---|
| L1: Tool-callable | APIs can be called by agents via MCP | Done |
| L2: Self-describing | Tools have descriptions and schemas | Done |
| L3: Identity-aware | Calls carry agent identity and ship info | Done |
| L4: Orchestratable | Tools can be composed into workflows | Done (WO + dispatch) |
| L5: Self-healing | Agents can retry and adapt on failure | Partial |
| L6: Learning | System improves from agent usage patterns | Planned |
Comparison with industry
| Platform | Agentic API approach |
|---|---|
| Anthropic | Claude API + Agent SDK + MCP |
| LangChain | Tool abstractions + Agent executors |
| SuperPortia | MCP servers + Dual agentic lines + WO system |
| AgenticAPI.com | Action-oriented API design patterns |
SuperPortia's approach is unique in combining MCP-based tool discovery with a work order system for task orchestration and a dual-engine architecture for cost optimization.
Related pages
| Page | Relationship |
|---|---|
| API Reference | Full API documentation |
| MCP Tools Overview | Tool catalog |
| Tool Discovery | How agents find tools |
| Engine-based Routing | Multi-engine dispatch |