---
title: "Discussion on \"Building a finance MCP connector: the simulation does the math, the LLM does the words\" | Hashnode"
description: "Discussion on \"Building a finance MCP connector: the simulation does the math, the LLM does the words\". Notes from DIVID Planner on keeping every number inside tool results, and on what Claude should say about them."
url: "https://hashnode.com/posts/building-a-finance-mcp-connector-the-simulation-does-the-math-the-llm-does-the-words/6afb6a1d9d7230a8956f6027"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Building a finance MCP connector: the simulation does the math, the LLM does the words](https://melnikov-ai.hashnode.dev/building-a-finance-mcp-connector-the-simulation-does-the-math-the-llm-does-the-words?utm_source=hashnode&utm_medium=feed)

Notes from DIVID Planner on keeping every number inside tool results, and on what Claude should say about them.

5d ago · 6 min read · melnikov-ai.hashnode.dev  
Read the article

-   [#mcp](https://hashnode.com/tag/mcp)
-   [#llm](https://hashnode.com/tag/llm)
-   [#typescript](https://hashnode.com/tag/typescript)
-   [#ai](https://hashnode.com/tag/ai)
-   [#software-architecture](https://hashnode.com/tag/software-architecture)

## Written by

[Ilya Melnikov](https://hashnode.com/@melnikov-ai)

@melnikov-ai

  
[View profile](https://hashnode.com/@melnikov-ai)

Founder of DIVID.AI. Simulation and planning software for people and brokerages.

## Responses 2

Markdown supported · ⌘↵ to post

[Ascendra Ventures](https://hashnode.com/@ascendra-ventures)

[5d ago](https://hashnode.com/posts/building-a-finance-mcp-connector-the-simulation-does-the-math-the-llm-does-the-words/6afb6a1d9d7230a8956f6027/comment/6afbbf6d6fe2f2d59e34b4b0)

"Every number comes from our model, and Claude's job is the words around it" is a clean rule, and the post shows how much work it takes to hold it. The fixed random seed per model version is the detail I liked most, because it means a difference between a plan and a what-if comes from the change and not from a fresh draw. Treating saved plan names as user text that can carry an instruction is also the right call, and an easy one to forget.

At Ascendra Ventures we reached the same boundary from a different direction while building memory for AI coding tools. Memory fixed the repetition, but not the trust: one parser was a single column off and reported numbers that were confidently, uniformly wrong. The lesson we wrote down was to put the model in the middle and never on the edges, and to decide up front which figure the model may propose and which figure plain code must be allowed to throw out. Your MCP server doing no planning math of its own is that boundary in practice.

Have you seen Claude paraphrase a likelihood word in a way that changed its meaning, and if so, did you fix it in the instructions or in the tool output?

[Ahmet Özel](https://hashnode.com/@ahmetozel)

[5d ago](https://hashnode.com/posts/building-a-finance-mcp-connector-the-simulation-does-the-math-the-llm-does-the-words/6afb6a1d9d7230a8956f6027/comment/6afb7198ba8669b3bad264c2)

For old cards, I would keep the historical result readable and replace the active controls with the run revision and a way to open the current result. That preserves the conversation's explanation without letting a past draft appear actionable.

The fixed seed also makes paired comparisons easier to reason about if both runs use the same scenario identities. I would preserve those identities with the market snapshot and show whether a comparison crosses model versions. Reusing the same numeric seed alone does not establish that two versions consumed the same simulated paths.
