Skip to content
Effloow
← Back to article
EFFLOOW LAB LAB-RUN

Microsoft Agent Framework 1 6 Anthropic Python Sandbox Poc 2026

Evidence notes document the bounded local or source-based checks behind an Effloow article. They are not product endorsements, legal advice, or benchmark claims.

Date: 2026-05-22 Content track: sandbox-poc Slug: microsoft-agent-framework-1-6-anthropic-python-sandbox-poc-2026

Goal

Verify that Microsoft Agent Framework 1.6.0 (the current stable release, originally announced as "MAF 1.0 GA") installs cleanly on macOS, exposes the Anthropic and Claude backends, and provides the advertised Workflow + step-based multi-agent API surface.

Environment

Host: macOS (Darwin 24.6.0)
Python: 3.12.x (system)
Sandbox venv: /tmp/effloow-maf-poc-1779451733/
Date: 2026-05-22
Secrets: none (no API calls made)

Installation

python3 -m venv /tmp/effloow-maf-poc
source /tmp/effloow-maf-poc/bin/activate
pip install agent-framework

Output (truncated):

Name: agent-framework
Version: 1.6.0
Summary: Microsoft Agent Framework for building AI Agents with Python.
Home-page: https://aka.ms/agent-framework
Author-email: Microsoft <af-support@microsoft.com>
Requires: agent-framework-core

Sub-packages installed (29 total)

Package Version
agent-framework-core 1.6.0
agent-framework-anthropic 1.0.0b260521
agent-framework-claude 1.0.0b260521
agent-framework-openai 1.6.0
agent-framework-foundry 1.6.0
agent-framework-a2a 1.0.0b260521
agent-framework-ag-ui 1.0.0rc3
agent-framework-devui 1.0.0b260521
agent-framework-orchestrations 1.0.0rc2

API exploration

import agent_framework as af

# Core workflow primitives
print(af.Workflow)        # <class 'agent_framework._workflows._workflow.Workflow'>
print(af.WorkflowAgent)   # <class 'agent_framework._workflows._agent.WorkflowAgent'>
print(af.FunctionalWorkflow)  # functional-decorator style
print(af.step)            # <function step at 0x...>

import agent_framework_anthropic as ant
print(ant.AnthropicClient)         # OK
print(ant.AnthropicBedrockClient)  # OK
print(ant.AnthropicVertexClient)   # OK

import agent_framework_claude as claude
print(claude.ClaudeAgent)          # OK
print(claude.ClaudeAgentOptions)   # OK

All imports succeeded without API keys.

Key findings

  • MAF 1.6.0 is the current PyPI release. The GA announcement used "1.0" as a branding milestone; the actual package version has been updated to 1.6.
  • Three Anthropic backends: AnthropicClient (direct API), AnthropicBedrockClient (AWS Bedrock), AnthropicVertexClient (Google Vertex).
  • agent_framework_claude provides ClaudeAgent — a native Claude-optimized agent without boilerplate.
  • agent-framework-devui ships a browser-based debugger for visualizing agent execution locally.
  • FunctionalWorkflow allows decorator-style workflow definition; Workflow uses an explicit graph-edge API.
  • WorkflowBuilder requires start_executor — not a plain constructor call; workflow creation goes through the @af.workflow decorator or explicit Workflow[T](start=step_fn) construction.

Limitations of this run

  • No actual API calls made (no API keys in sandbox).
  • WorkflowBuilder.__init__() requires start_executor kwarg — full constructor test was not run.
  • DevUI startup not tested (requires running process).

Verdict

Install: PASS. API surface: PASS. Anthropic/Claude backends: PASS. Sufficient evidence for a practical PoC article targeting developers who want to migrate from AutoGen or Semantic Kernel to MAF 1.6.0 with an Anthropic backend.

Read the article

This note supports the public article and records what was actually checked.

Open article →