01
Discover
Search bundled knowledge, inspect FROOT modules, and look up AI terminology offline.
Knowledge + glossary
Python-native delivery
FrootAI for Python
Use the SDK in your code. Use Python MCP when an agent needs the tools.
The same FAI Engine in two Python-native forms: direct APIs for applications and notebooks, or a local MCP server that gives AI agents an inspectable, trust-gated tool surface.
Python 3.10+Wheel + source distributionMIT licensedSDK 5.1.0 · MCP 6.1.0
python · frootaiSDK 5.1.0
from frootai import FrootAI, SolutionPlay, Evaluator
client = FrootAI()
results = client.search("RAG architecture")
module = client.get_module("R2")
play = SolutionPlay.get("01-enterprise-rag")
cost = client.estimate_cost(play.id, scale="prod")
# Keep the workflow explicit and inspectable.Verified release
A compact runtime. A much larger system.
SDK 5.1.0 and Python MCP 6.1.0 package the current FAI Engine with its operating boundaries visible.
62
MCP tools
Knowledge, evaluation, scaffolding, trust, and federated routing
101
Solution Plays
Packaged reference architectures from starter to enterprise
863+
FAI primitives
Agents, skills, instructions, hooks, and plugins
0
SDK dependencies
A portable standard-library core for Python 3.10+
Choose the process boundary
SDK or Python MCP?
Choose by who makes the call. Your Python application imports the SDK; an external AI client launches the MCP process.
frootai 5.1.0
Python SDK
Best for notebooks, Python applications, evaluation code, and automation that needs direct return values.
pip install frootaiOpen SDK on PyPI frootai-mcp 6.1.0
Python MCP Server
Best for VS Code, Claude, Cursor, or another MCP client that should call FrootAI tools through a local Python process.
pip install frootai-mcpOpen Python MCP on PyPI Inside the FAI Engine
From lookup library to decision system
Python carries the connected workflow: discover evidence, compare options, build with guidance, evaluate quality, and federate trusted tools.
02
Decide
Compare models and Solution Plays, inspect architecture patterns, and estimate Azure cost direction.
Models + plays + cost
03
Build
Move through the FAI build → review → tune chain with explicit guidance at each stage.
Agent workflow
04
Evaluate
Check configurable quality thresholds and run A/B experiments with callbacks you control.
Quality + experiments
05
Connect
Attach external MCP areas through trust policy and invoke their tools through one Python process.
Trust-gated federation
The product map
Know what each package owns
Knowledge access
Search FROOT modules, inspect architecture guidance, and look up AI terms.
Solution Plays
Browse the packaged Play catalog and load a known architecture contract.
Cost and evaluation
Estimate directional cost and check evaluation thresholds from Python.
Notebook friendly
Use the same API from scripts, services, notebooks, and automation.
Agent-facing tools
Expose FrootAI capabilities over MCP instead of importing the SDK into the client.
Stdio process
Launch the Python server as `frootai-mcp-py` from a compatible MCP client.
Local process boundary
Keep the Python runtime and its environment under your control.
Python core access
Use `FrootAIMCP` directly when embedding the server core in Python.
The working loop
Search → inspect → evaluate → integrate
The package does not make architecture decisions invisible. Your code still chooses the query, Play, scale, and thresholds.
01
Search
Query bundled FROOT knowledge from the process.
02
Inspect
Open a module or compare a Solution Play before acting.
03
Evaluate
Check cost direction and quality thresholds in code.
04
Integrate
Keep it in-process or expose it to an agent through MCP.
Start with a clean environment
Two verified setup paths
Python SDK
python -m venv .venv# Activate .venv for your shellpython -m pip install --upgrade pippip install frootaipython -c "from frootai import FrootAI; print('FrootAI ready')".vscode/mcp.json
{ "servers": { "frootai-python": { "type": "stdio", "command": "frootai-mcp-py" } }}Backend, made visible
Local by default. Live by choice.
The useful core travels with the wheel. Network integrations sit at the edge, and every federated attach passes through explicit trust policy.
Caller
Python app or AI agent
Direct SDK calls or protocol calls through MCP.
→
Core
FAI Engine
Search, Plays, cost, evaluation, scaffold, and build-review-tune.
→
Evidence
Bundled + federated
Offline catalogs first; trusted MCP areas and live sources on demand.
Ready boundary
Python 3.10+. Isolated environment. Explicit process choice.
The SDK stays in your application process. Python MCP runs as a separate stdio server. Hosted and upstream integrations retain their own network requirements.