Add AER-1 receipts to a Pydantic AI agent: check what it ran, not what it says it ran.

Route 1: the direct MCP pattern

Pydantic AI speaks MCP natively, so the shortest path needs no Zambo package at all. Register the Zambo endpoint as an MCP server toolset and the agent gains Zambo's tools (153 on the full surface, 35 loaded by default) alongside its own. Calls execute on the server, and each one leaves a receipt because the executing layer makes the record.

from pydantic_ai import Agent
from pydantic_ai.mcp import MCPServerStreamableHTTP

zambo = MCPServerStreamableHTTP('https://zambo.dev/mcp')  # no API key, no account
agent = Agent('openai:gpt-4o', toolsets=[zambo])

result = agent.run_sync('Check the live price of ETH and save what you find.')
# Every Zambo tool call in that run returned a verifiable AER-1 receipt.

The free tier covers 20 free calls per day, so a first Pydantic AI agent can run receipted today with no payment. The per-client connect steps for every MCP host live on zambo.dev/add, and the pattern behind this route is written up in MCP server receipts.

Route 2: wrap your own tools with aer1-pydantic

Route 1 receipts the calls your agent makes to Zambo's tools. If the work that matters runs through your own Pydantic AI tools, wrap those calls instead: aer1-pydantic on PyPI wraps the agent's tool calls so each one returns an AER-1 receipt as it runs, in the same format and checkable the same way.

pip install aer1-pydantic

Both routes produce the same object: one receipt per tool call, recording the tool that ran, a hash of the arguments it received, the exact result bytes and their SHA-256 digest, and a timestamp, published at a public URL. A run that mixes Zambo tools and wrapped local tools therefore checks as one record, not two systems.

A worked example: one real receipt

This is a real receipt from the public Zambo receipt store, opened and checked while this page was written: receipt d6949ef2-d6f7-46ec-a902-cf8be1ebeb7a, tool memory_remember, saved 2026-10-25T14:41:56.661Z, output hash sha256:716f3d7dc745c642ab467e4c999bc8a918e990c1abce72607393a476093aeca1. Its public page states the result has not changed since it was saved, and its commitment check reads: commitment matched.

• Read it: the receipt's public run page shows the saved result, the canonical bytes, and the checks.
• Check it: paste the receipt ID into the verifier at zambo.dev/verify, or call the public verify endpoint and compare the returned output hash with the one above. A match proves the saved result has not changed since it was saved.
• Yours looks the same: a receipted Pydantic AI call returns this same object shape, so the check you just ran is the check your users, clients, and reviewers will run on your agent's work.

Checking the run afterwards

• Per call: open the receipt URL and recompute the commitment from its canonical bytes, by hand or with the live verifier. A match means the saved record is the record made when the call ran.
• As an audit: the receipts are the audit trail, countable and checkable after the fact. The method is written up in how to audit AI agent tool calls, and the verifier walkthrough runs the procedure against a real public receipt.
• Full days covered: a Day Pass ($0.99) keeps every call receipted for 24 hours, past the 20 free calls per day on the free tier.