I'm working on leveraging NLP and LLM techniques to create a geometry over the discrete space of Ethereum transaction execution structure. (sorry... it's a bit of a mouthful)
Fine tuning Gemini 2.5 flash on EVM Blockchain execution structures and using perplexity with PageRank to perform transaction anomaly detection.
https://lite.chaingenius.ai/theory
Odd seeing this right now for me. I recently implemented a 6NF schema for parsed XBRL files from EDGAR. The architecture was the right call... too bad the data is not useful for analytics.
I moved from pip to poetry a while back, but was introduced to uv recently while working on MCP stuff. I now use uv for everything and haven’t looked back.
Just a fancy way of saying that multiple agents (with their own LLMs) can concurrently connect, also that pg-mcp can simultaneously connect to multiple Postgres servers as well.
The use case for pg-mcp is server deployment - local running is just for dev purposes. HTTP/SSE enables multiple concurrent connections and network access, which stdio can't provide.
Xata.io looks very interesting!!! I was thinking about building an intelligent agent for pg-mcp as my net project but it looks like you did a lot of the hard work already. When thinking about the "AI Stack" I usually separate concerns like this:
I built this to fill a specific need. It's not really made to talk to Claude Desktop (although it can). It's built to take multiple connections from agents who bring their own LLM (via API key or whatever) and provide context to the LLM model. Yes, it does import the standard Python MCP package, however it does quite a few non-standard things in order to achieve the multi-tenancy. I encourage you to look at the code.
The pg-mcp server doesn’t need an API key. Those variables are only needed to run the example claude-client. You can write your own agent to talk to whatever LLM you like.
It won’t be that easy. First off, most databases in the wild are not well documented. LLMs benefit from context, and if your tables/columns have non-intuitive or non-descriptive names, the SQL may not even work. Second, you might benefit from an LLM fine-tuned on writing code and/or an intelligent Agent that checks for relevancy and ambiguity in user input prior to attempting to answer the question. It would also help if the agent executed the query to see how it answered the user’s question. In other words “reasoning”… pg-mcp simply exposes the required context for Agents to do that kind of reasoning.
Personal story: I sat in his office as an undergrad and he presented a problem to me concerning counting balls in boxes that had applications to quantum theory. It blew my freshman mind.
1) At the time, the European data laws implied that it protected its citizens no matter where they are. Nobody wanted to be the first to test that in court.
2) The organizations and agencies performing this type of data modeling were often doing so on behalf of large multinational organizations with absurd advertising spends, so they were dealing with Other People’s Data. The responsibility of scrubbing it clean of EU citizen data was unclear.
What this meant was that an EU tourist who traveled to the US, and got served a targeted ad, could make a RTBF request to the advertiser (think Coca-Cola, Nestle or Unilever)
Agreed. The media and advertising industry was most definitely leveraging cookie-level data for building attribution and targeting models. As soon as the EU established that this data was “personal data”, as it could, theoretically, be tied back to individual citizens, there were questions about the models. Namely “Would they have to be rebuilt after every RTBF request?” Needless to say, no one in the industry really wanted to address the question, as the wrong answer would essentially shut down a very profitable practice.
https://www.chaingenius.ai
The goal is to find on-chain structural anomalies, as well as seeing if clustering by behavior has emergent semantic properties