LangChain setup for FrankKi

Setup

python -m pip install langchain-mcp-adapters langgraph

import os
from langchain_mcp_adapters.client import MultiServerMCPClient

client = MultiServerMCPClient({
    "frankki": {
        "transport": "streamable_http",
        "url": "https://mcp.frankki.app",
        "headers": {"Authorization": f"Bearer {os.environ['FRANKKI_API_KEY']}"},
    }
})
tools = await client.get_tools()

1. 1.Filter the returned tools before binding them to a model. Start with mcp_health, address_validate, letter_create_draft, letter_preview, shipping_quote, and order_status.
2. 2.Use a LangGraph interrupt after quote and before any submission. Resume only from an external human decision bound to the reviewed artifact and quote.
3. 3.Route PRICE_CHANGED back through quote and a fresh interrupt. Preserve clientOrderId through bounded retries.
4. 4.Do not checkpoint recipient data or letter content unless the checkpoint store has an approved retention and access policy.

Required FrankKi safety boundary

Validate the address, create a stable draft, inspect every preview page, obtain a fresh quote, set maxCostEuros, persist a stable clientOrderId, and obtain an explicit human decision before a live submission. Framework approval controls supplement FrankKi approval; model text never substitutes for the human decision.

Continue with canonical agent onboarding, the draft guide, send guide, and error recovery.