Dazza Greenwood dazzaji

Backgrounder: Grasshopper Bank AI Agent MCP Server

Grasshopper Bank announced in August 2025 that it now offers a Model Context Protocol (MCP) server, enabling some account holders to interact with their accounts using AI Agents and API-based workflows. This represents a major step in digital banking for small businesses, allowing secure conversational access to account data through integrations such as Claude by Anthropic.[1][2][3]

What is the MCP Server?

The MCP server is a backend protocol designed to securely standardize and orchestrate how AI agents interact with financial account data. It enables business clients to request personalized financial analysis, transaction categorization, budgeting advice, and real-time cash flow insights via AI assistants, starting with Claude and potentially expanding to other LLMs in the future.[4][5][2][6][1]

How Does It Work?

Tobin,

Here’s a compact suggested plan that others can lift into production after Monday. It involves demo of two portable artifacts:

Grant of Authority — what the user (Principal) authorizes an agent to do, with purpose‑binding, explicit capabilities + constraints, negative constraints (exclusions), time bounds, data‑minimization categories, revocation, and a verifiable token counterparties can check.
Ask/Confirm Policy — how much autonomy the user wants and the exact circumstances that trigger permissions or clarifications (presets up front; customize later).

Why this lands: it converts apparent authority into provable authority, and makes “duty‑of‑loyalty‑like” behavior observable (authenticating requesters, selective disclosure, audit you can read). It also sets up safe portability across agent providers.

PROMPT

Can you turn this into a markdown table? Or into CSV. Tell me if you’re unable to make sense of all the data presented.

attached image

OUTPUT (formatted)

I think congress and the new administration should craft a modern statutory prong of fair use explicitly tailored to AI training data access, focused specifically on models that provide free/open access tiers or are open source. This targeted approach recognizes that existing copyright frameworks, rooted in decades-old standards, fail to address a transformative technology that is critical to economic growth, societal advances, and the greater good. While these models use copyrighted works in their training, they do so in ways that are highly innovative and do not directly compete with the original works. The carve-out could be part of a broader "new deal on data" that pairs expanded fair use protection with new obligations around transparency and limited licensing requirements - creating a balanced framework that promotes continued advancement in this essential field while providing appropriate safeguards and accountability.

Interested to learn more and consider this? I asked OpenAI Deep Research to help

The following is a longer reply to a comment than LinkedIn will permit.

Thanks for asking! I wrote “nope” because I fundamentally reject the idea that having AI significantly smarter than humans automatically poses an inherent problem. In fact, I see it as an exciting opportunity.

Here’s why:

Tools Don’t Automatically Dull Our Minds: Historically, new technologies—like calculators—haven’t led to a wholesale decline in human intelligence. Instead, they shift how we use our cognitive resources. Rather than seeing AI as replacing human problem-solving, I view it as freeing us to explore more creative, strategic, or empathetic forms of thinking.

What is NotebookLM

NotebookLM gives you a personalized AI collaborator that helps you do your best thinking. After uploading your documents, NotebookLM becomes an instant expert in those sources so you can read, take notes, and collaborate with it to refine and organize your ideas.

NotebookLM is available to all users (18+) in the 180+ regions where the Gemini API is available in.

Asking questions

You can interact with your sources by writing questions or instructions into the chat box. Click the arrow or type the Enter key to send your question.

Things to try:

This gist is just me trying to better understand Ruv's idea for "Hypergraphs" as a way to do cool stuff with data (especially through generative AI and including for legal contexts). Here's more color on how Hypergraph's can be applied to contracts: https://gist.github.com/ruvnet/4dde5fcbaf13c88678e63ab75ff9bf78

In this gist there are three items, Section A, immediately below, is a hypothetical UCC Article 2 type contract for the sale of goods (a widget) between Alice (seller) and Bob (buyer) to explore the hypergraph's potential for capturing complex legal relationships and timelines. It is followed by Section B) Hypergraph Specification for Contract Between Alice and Bob, and Section C) Annotated Application of Hypergraph to the Contract.

A. Sale of Widget Agreement

This Sale of Goods Agreement ("Agreement") is made and entered into as of March 13, 2024, by and between Alice Ace ("Seller"), a legal entity specializing in manufacturing widgets, located at 1 Vericilious Lane, Berkington, CA 95678, a