LangChain Tools

Runnable example: 27_integrations_mcp_langchain_dbt.ipynb — all three integrations end-to-end, offline, executed with outputs.

Expose Portiere's operations as LangChain tools for use in agents and chains. Works with langchain-core alone. All tools run offline.

pip install "portiere-health[langchain,polars]"

from portiere.integrations.langchain import get_langchain_tools

tools = get_langchain_tools()          # list[StructuredTool]

# e.g. a ReAct agent
from langchain.agents import create_react_agent   # your LLM setup
agent = create_react_agent(llm, tools, prompt)

Tools

Same five operations as the MCP server (they share one core): portiere_list_standards, portiere_profile_source, portiere_suggest_schema_mapping, portiere_map_concepts, portiere_egress_posture. Each returns a JSON string an agent can parse.

Direct invocation

tools = {t.name: t for t in get_langchain_tools()}

print(tools["portiere_list_standards"].invoke({}))
print(tools["portiere_suggest_schema_mapping"].invoke(
    {"columns": [{"name": "diagnosis_code"}], "target_model": "omop_cdm_v5.4"}))
print(tools["portiere_egress_posture"].invoke({}))   # verify no-egress

Safety

Every tool constructs Portiere with offline=True; a remote-provider config is rejected before any network call. portiere_egress_posture lets the agent (or your guardrail) confirm the posture programmatically.