tai-daf-sdk 2.0.0a2

Python SDK for DAF (Declarative Agentic Framework) — thin MCP client

DAF SDK

Python SDK for DAF (Declarative Agentic Framework) — a platform for building, managing, and orchestrating AI agents and multi-agent teams.

SDK 2.0 (SDK-MCP-R1 + R2) is a thin MCP client with a codegen-typed surface: client.<resource>.<method>() maps to daf_<method>_<resource>, returns Pydantic models when the MCP tool has an output_schema, and stays in sync via catalog drift CI.

Features

• Thin MCP dispatch — Any client.<resource>.<method>(**kwargs) → daf_<method>_<resource> (no per-resource SDK modules)
• Typed returns (R2) — Generated models.py / _tool_registry.py wrap tool results; unknown tools fail fast
• IDE stubs — Generated types.py TypedDicts + client.pyi for kwargs autocomplete
• Sync & Async — DAF and AsyncDAF share the same public surface
• Streaming — client.stream(tool_name, args) yields MCP notification/progress events
• Catalog sync CI — Daily drift check opens a PR when the backend MCP catalog changes

Requirements

• Python 3.10+
• Running DAF backend with MCP endpoint ({base_url}/mcp)

Installation

pip install tai-daf-sdk

Or install from source:

git clone https://github.com/your-org/tai-daf-sdk.git
cd tai-daf-sdk
pip install -e ".[dev]"

Quick Start

from tai_daf_sdk import DAF
from tai_daf_sdk.models import ListAgentsResponse

with DAF(base_url="http://localhost:8012", api_key="daf_your_api_key") as client:
    # → MCP tool daf_list_agents; ListAgentsResponse when output_schema exists
    agents = client.agents.list()
    assert isinstance(agents, ListAgentsResponse)
    print(agents)

    # Stream a long-running tool (manual API — not codegen'd)
    for event in client.stream("daf_test_agent", {"name_or_id": "assistant", "sample_input": "Hi"}):
        print(event)

Async

from tai_daf_sdk import AsyncDAF

async with AsyncDAF(base_url="http://localhost:8012", api_key="daf_...") as client:
    agents = await client.agents.list()

Authentication

# Explicit API key
client = DAF(base_url="http://localhost:8012", api_key="daf_...")

# Or set DAF_API_KEY in the environment (resolved by MCP transport)

Optional advanced: set DAF_MCP_STDIO_COMMAND to use stdio MCP instead of HTTP.

Regenerating typed surface (do not hand-edit)

Generated files (tai_daf_sdk/models.py, types.py, _tool_registry.py, client.pyi) are codegen-only:

python -m tai_daf_sdk.codegen \
  --mcp-url https://tai-daf-sdk-backend-dev.azurewebsites.net/mcp \
  --api-key "$DAF_API_KEY" \
  --out tai_daf_sdk/

Offline / tests:

python -m tai_daf_sdk.codegen --catalog-json tests/fixtures/mcp_list_tools_golden.json --out tai_daf_sdk/

IDE autocomplete uses tai_daf_sdk/client.pyi + TypedDicts in tai_daf_sdk/types.py.

1.x → 2.0 note

1.x:  agents = client.agents.list()   # list[Agent]
2.0:  result = client.agents.list()   # ListAgentsResponse (or dict if no output_schema)

REST mode and hand-maintained resource modules are removed from 2.0 (see 1.x-maintenance branch). Full migration guide arrives in R3.

Development

pip install -e ".[dev]"
pytest tests -q
# Live MCP (optional):
# pytest -m mcp -v

License

MIT