SQL AI & MCP

Agentic AI Kit extends traditional Python workflows with SQL-powered AI enrichment, lightweight multi-agent orchestration, and Model Context Protocol (MCP) integration. These components are designed for notebook workflows, data science, automation, and agent ecosystems.

SQL AI

SQL AI allows Gemini to generate new columns directly from SQLite query results.

Instead of exporting rows into a separate prompting workflow, AI generation becomes part of the SQL query itself.

Typical use cases include:

Text summarization
Automatic classification
Keyword extraction
Retrieval-friendly document generation
Data enrichment
Analytics pipelines

Example Workflow

import sqlite3import pandas as pd
from agentic_ai.sql_ai import run_ai_sql

conn = sqlite3.connect(":memory:")
records = pd.DataFrame(    {        "record_id": [1, 2, 3],        "topic": [            "Access",            "Documentation",            "Automation",        ],        "notes": [            "User cannot access account.",            "Setup guide needs examples.",            "Messages should be classified.",        ],    })
records.to_sql(    "records",    conn,    index=False,)
sql = """SELECT    record_id,    topic,    notes,    ai_generate(        prompt='Write a retrieval summary'    ) AS retrieval_summaryFROM records"""
result = run_ai_sql(    sql=sql,    conn=conn,)
print(result)

Supported SQL AI Functions

ai_generate()

Generate new text from each row

ai_summarize()

Produce concise summaries

ai_classify()

Assign predefined labels

ai_extract()

Extract keywords or entities

These functions execute after the relational query while preserving the familiar SQL workflow.

Writing Results Back to SQLite

SQL AI also supports notebook-friendly table creation.

CREATE OR REPLACE TABLE enriched_records AS
SELECT    record_id,    notes,
    ai_summarize(        model="gemini-2.5-flash"    ) AS summary,
    ai_extract(        prompt="Extract keywords",        count=5    ) AS keywords
FROM records;

run_ai_sql() transparently handles the table creation and writes the enriched dataframe back into SQLite.

Multi-Agent Patterns

Agentic AI Kit includes lightweight orchestration patterns for coordinating multiple agents.

Conversation

Round-robin discussion between multiple agents.

run_conversation(...)

Supervisor

Delegate work from one coordinating agent to specialist agents.

run_supervisor(...)

Parallel Execution

Run multiple agents simultaneously.

run_parallel(...)

Debate

Generate opposing viewpoints and allow a judge agent to evaluate them.

run_debate(...)

These patterns provide reusable orchestration primitives while remaining lightweight and transparent.

Model Context Protocol (MCP)

The MCP package enables Agentic AI Kit to expose and consume tools over HTTP using an MCP-style architecture.

Components

MCPServer

Publish tools over HTTP

MCPClient

Discover and invoke remote tools

This enables distributed tool execution and interoperability between independent agent systems.

The current implementation can be explored in:

agentic_ai/examples/11_mcp_demo.py

Runnable Examples

Agentic AI Kit ships with a growing collection of runnable examples covering the major framework capabilities.

01_talking_agents.py

Round-robin conversations

02_personality_and_memory.py

Personas and conversational memory

03_tool_agent_weather.py

Gemini function calling

04_json_agent.py

Structured JSON output

05_reasoning_agent.py

Planner-to-executor workflow

06_supervisor.py

Supervisor delegation

07_parallel_agents.py

Parallel agent execution

08_debate.py

Debate and judging

09_long_term_memory.py

Persistent SQLite memory

10_rag_agent.py

Retrieval-Augmented Generation

11_mcp_demo.py

MCP server and client

12_emotion_memory.py

Emotion-aware episodic memory, replay, tree attention, and reinforcement learning

Running an Example

Execute any example from the repository root.

python agentic_ai/examples/01_talking_agents.py
Putting It All Together

Agentic AI Kit combines conversational agents, memory systems, retrieval, structured outputs, SQL AI, MCP integration, and multi-agent orchestration into a unified Python framework. Whether you're building analytical copilots, autonomous workflows, retrieval systems, or adaptive research agents, the framework provides modular components that can be composed into larger agent architectures.

Page updated

Google Sites

Report abuse