Open Source • MIT Licensed

Agent As
Everything

Your agent is a tool. Your agent is a MCP server. Your agent is a function. With AA Kit, agents transcend categories.

quick_start.py
LIVE
from aakit import Agent
# Create agent
agent = Agent(
name="assistant",
instruction="You are a helpful assistant",
model="gpt-4"
)
# Instant MCP server
agent.serve_mcp()
MCP Server Ready
3 lines → Production ready

Agent As

In AA Kit, agents aren't just MCP servers-they're tools, functions, and services. One primitive, infinite forms.

mcp_composition.py
Create Agent
# Step 1: Create your agent
agent = Agent(
  name="researcher",
  instruction="You research topics thoroughly",
  model="gpt-4"
)
MCP Compatible • Composable • Production Ready
“When you realize an agent can be a tool for another agent, and that toolis also an agent, you understand: agents are the universal primitive of AI.”
Zero Config
Universal Protocol
Instant Server

Why Agent As?

Traditional frameworks box agents into roles. AA Kit sees agents as the universal primitive-capable of being anything.

Agent as Server

Every agent can serve itself via MCP, making it accessible to any compatible tool or platform.

Agent as Tool

Agents can be passed as tools to other agents, enabling natural composition and delegation.

Agent as Function

At its core, an agent is a pure function-deterministic, composable, and infinitely scalable.

“One agent, many forms. That's the AA philosophy - agents aren't confined to categories, they transcend them.”

3
Lines to Production
0
Config for MCP
Composition Depth
1
Universal Protocol

The AA Philosophy

Agent As: One primitive, many forms. See how the same simple pattern adapts to every use case.

Agent = Function

Three parameters: name, instruction, model. Returns intelligence. No classes, no inheritance, no complexity.

Agent = Server

Every agent exposes itself as an MCP server instantly. No setup, no configuration-just agent.serve_mcp().

Agent = Tool

Pass one agent to another as a tool. They compose naturally because they're all the same primitive underneath.

Agent = Production

Rate limits, retries, caching, circuit breakers. Not added later-built into the constructor from day one.

Agent = Any Model

OpenAI, Anthropic, local models. The agent abstraction remains constant; only the implementation changes.

Agent = Stateless

Pure functions with injected memory. Horizontally scalable by design, not by accident.

Progressive Complexity

The same three-line pattern scales from prototypes to production. Complexity is optional, simplicity is not.

example.py
agent = Agent(
    name="assistant",
    instruction="You are a helpful assistant",
    model="gpt-4"
)
# Simple synchronous usage - no async needed!
response = agent.chat("Hello!")

# Or use async when you need it
response = await agent.achat("Hello!")

Compare AA Kit

See how AA Kit stacks up against other popular frameworks.

Feature
X
AA Kit
ADKAutoGen
Setup Complexity3 lines20+ lines15+ lines
MCP SupportNativeVia toolsNot supported
Production FeaturesBuilt-inManual setupLimited
Agent CompositionNaturalComplexConversation-based

The Future is Composable

Join the developers building the next paradigm of AI systems.