[Live] Le MCP Finary est disponible, construisons-le ensemble

Finary Announcements
84

Hello @Mariya_Finary

The MCP is really useful for AI-powered analysis.

On my end, I have entered the management fees for my life insurance and PER (retirement savings plan) contracts in Finary. However, this information does not seem to be returned by the MCP connector.

Are you planning to expose via the MCP, for each contract:

  • the annual fees on the euro fund;

  • the annual fees on unit-linked investments;

  • the additional fees for managed or discretionary management;

  • the fees on contributions and transfers;

The MCP currently returns the internal fees of certain investment vehicles, such as the expense ratio of funds and ETFs, but not the fees specific to the wrapper. This data would be particularly useful for calculating the actual total cost of each contract and comparing investments.

Thank you.

Christophe

2 Likes
85

Hello @Mariya_Finary,

I suggest enhancing the transaction search functionality via the MCP.

The current search_spending function allows searching for a label and obtaining the total amount and number of occurrences:

{
  "tool": "mcp__Finary__search_spending",
  "arguments": {
    "query": "Amazon",
    "direction": "spending",
    "period": "year_to_date"
  }
}

It would be interesting to add two complementary functions.

1. search_spending_list

Obtain all labels present over a period, with their number of occurrences and cumulative amount.

{
  "tool": "mcp__Finary__search_spending_list",
  "arguments": {
    "direction": "spending",
    "period": "year_to_date"
  }
}

Expected response:

{
  "items": [
    {
      "label": "AMAZON EU",
      "occurrences": 6,
      "total": "320.40"
    },
    {
      "label": "CARREFOUR",
      "occurrences": 4,
      "total": "202.10"
    }
  ]
}

This would allow an AI to automatically identify the main merchants/labels without having to know their names in advance.

2. search_spending_details

Obtain the details of individual transactions matching a search.

{
  "tool": "mcp__Finary__search_spending_details",
  "arguments": {
    "query": "Amazon",
    "direction": "spending",
    "period": "year_to_date"
  }
}

Expected response:

{
  "matched_transactions_count": 2,
  "matched_expense_total": "420.80",
  "transactions": [
    {
      "label": "AMAZON EU",
      "date": "2026-10-01",
      "amount": "30.40"
    },
    {
      "label": "AMAZON EU",
      "date": "2026-09-02",
      "amount": "390.40"
    }
  ]
}

We would thus have three complementary levels:

  • search_spending_list → list of labels + occurrences + totals
  • search_spending → total + number of occurrences for a specific label
  • search_spending_details → individual transactions for a specific label

This would enable much more advanced analyses via the MCP: recurring expenses, main merchants, period comparisons, searching for unusual transactions, etc.

1 Like
86

@Daniel62 @user32 @Dimitri13 the error on outputSchema (tools 3 and 4) has been fixed since September 22nd. Can you retry the connection and let us know if the tools are working properly?

87

Hello,
I’ve isolated a rather peculiar behavior with the Finary MCP in ChatGPT’s Scheduled Tasks.
Reading from Finary works correctly, but as soon as Finary has been called in the task, a subsequent write to Google Sheets is blocked by ChatGPT with:
This tool call was blocked by OpenAI’s safety checks. Please double check what you are sending.
I’ve performed several A/B tests with the same Google Sheet:
Google Sheets alone, including with the same financial amounts → :white_check_mark:
CoinGecko → neutral string → Google Sheets → :white_check_mark:
Gmail → neutral string → Google Sheets → :white_check_mark:
Finary → Finary data → Google Sheets → :cross_mark:
Finary → data completely ignored → writing a simple constant TEST_AFTER_FINARY_2026-10-07 → Google Sheets → :cross_mark:
The last test seems particularly interesting to me: no data from Finary is passed to Google Sheets. The task reads Finary, completely ignores the results, then attempts to write only a hardcoded string from the prompt. The write is nonetheless blocked by OpenAI’s safety checks.
Equivalent tests after reading from Gmail or CoinGecko work, including writing, reading back, and clearing the cell.
Additionally, my Scheduled Task Finary → Google Sheets workflow worked previously.
Do you know if the Finary MCP is declared or classified in a particular way by ChatGPT/OpenAI, especially given that it provides access to personal financial data?
Are you aware of any restriction or recent change on OpenAI’s side regarding the use of another connector after a call to the Finary MCP?
If needed, I can provide the exact test protocol and the prompts used.

88

Hello @plschaming_Finary,
It’s working again on LM Studio. Thank you very much for your work.

89

Yes, it’s usable.

I’ve been doing it since the beta.

So now that it’s the final version, you can 100% use it with LM Studio and Qwen or other local models.