The Fincome MCP lets you query your KPIs in natural language from your AI assistant — and turn those answers into real deliverables. This article explains who it is for, how it works, what information it can return, and shows concrete examples built end to end.

Who is the Fincome MCP for?

The MCP doesn't replace the dashboard: it saves you from opening it for every figure, and above all it lets you go straight from data to deliverable. What it brings depends on your role:

How does it work?

The principle is simple: your AI assistant becomes the pilot, Fincome remains the source of truth.

1. You ask a question in natural language in your assistant (Claude, Cursor…), for example: « Break down my ARR growth over the last 12 months by movement type. »
2. The assistant translates the question into a query to Fincome: it picks the metric, the period, the time granularity (month, quarter, year) and the breakdown dimension.
3. Fincome returns the calculated figures — exactly the ones that feed your dashboard, with the same calculation options, the same recognition rules and the same scopes.
4. The assistant formats the answer: table, summary, chart, Excel file, presentation. This is where the value is created — you don't get raw data back, you get a deliverable.

Three points to keep in mind:

• It is not a chatbot built into Fincome. It is your assistant fetching the data from us. You stay in your own working environment, alongside your other connected sources.
• Access is read-only. The assistant can consult everything your permissions allow, but can change nothing in Fincome.
• The figures are not recalculated by the AI. They come from Fincome. However, if you ask the assistant for a derived calculation (a ratio, a projection, a weighted average), that calculation is its own — and should be checked like any spreadsheet formula.

For the setup, see Connect the Fincome MCP.

What information can the MCP return?

The MCP gives access to your aggregated KPIs and your refined detailed views. In practice:

What the MCP does not cover yet

The MCP's scope is narrower than the dashboard's. The following are excluded for now:

• cash in, cash out and cash flow variation;
• the Forecast (forecasting models and scenarios);
• the MRR by customer and price view (mrr_by_customer_and_price);
• the raw objects from your billing sources (invoices, customers and subscriptions as imported).

Three examples of deliverables built with the MCP

Here are three real deliverables, built on a demo account, purely by asking questions in natural language — with no exports, no SQL and no manual reprocessing.

Example 1 — A financial review datapack (due diligence style)

What it is: an 11-tab Excel workbook and a 14-slide presentation answering the questions an investment fund would ask about the company. Production time: about ten minutes, 11 questions.

It covers:

• the ARR breakdown by industry, account size, acquisition channel and product;
• recurring vs one-off revenue and deferred revenue levels;
• collection by invoice status and the unpaid rate;
• average revenue per account (ARPA) and LTV over the last 12 months, overall and by segment;
• the growth breakdown (new / reactivation / expansion / contraction / churn), including the mix effect;
• acquisition by sales rep, upsell and NRR by CSM, and 24-month NRR cohorts;
• a summary scorecard of SaaS KPIs (ARR, YoY growth, NRR, GRR, quick ratio, churn, recurring share).

Some of the questions used to produce it:

• « What are the main drivers of ARR growth: price, volume or new customers? »
• « Is the company exposed to customer concentration risk? »
• « Which segment (account size / geography) is driving growth the most? »
• « What are the main reasons for churn? Is it more pronounced on certain products or customer types? »
• « What analysis would a tech investment fund run on this company? »

Downloads: the presentation (PDF) · the Excel workbook

Example 2 — A 36-month ARR projection

What it is: a 10-page FP&A working document projecting ARR over three years, in three scenarios (conservative, central, ambition), with the expected-variation bridges and sensitivities by driver.

The deliverable details:

• the starting point: the ARR trajectory and the bridge of the last 12 months (new business, expansion, reactivation, contraction, churn, FX effect);
• the methodology and assumptions: a driver-based model calibrated on the last 12 months, split into three explicit scenarios, with the model's limitations;
• the three ARR trajectories at 12, 24 and 36 months, with the associated CAGR;
• the expected-variation bridge, year by year;
• the sensitivities: what 1 point of gross churn, 1 point of expansion or 10 points of new business growth is worth;
• the projections broken down by account size, geography and product tier.

The key point: the model is calibrated on Fincome's native bridges, not on arbitrary assumptions. The churn, expansion and reactivation rates per segment come straight from your historical data. Note that this model is built by the assistant from history — it is not Fincome's Forecast module, which is not accessible through the MCP.

Downloads (French version): the projections document (PDF) · the Excel model

Example 3 — Answers on the fly

This is the most everyday use, and often the most useful. No formatted deliverable — just an immediate answer in the flow of your conversation.

• « How many new customers in May? »
• « What is my churn rate this quarter? »
• « Show me my monthly MRR over the last 12 months. »
• « What is my ARR by product? »
• « Compare my MRR in January and June, and explain the gap. »
• « Which customers downgraded this quarter? »

Best practices for getting good answers

1. Always specify the period. « Over the last 12 months », « from 1 January to 31 March 2026 », « as of the last day of the quarter ». This is the number one source of gaps between what you expect and what you get.

2. Name the breakdown dimension. « By product », « by country », or the exact name of your analytical dimension. Beyond product and country, the assistant can only break figures down along dimensions you have configured.

3. Ask the assistant to cite its figures and reconcile them. For example: « Check that the sum of the segments equals the consolidated total, and flag any gap. » This is the best safeguard on complex analyses.

4. Work in iterations. Ask a broad question, read the answer, then refine. The best deliverables are built from 10 chained questions, not from one very long request.

5. Ask explicitly for the output format. « Put this in a table », « generate an Excel file with one tab per analysis », « build me a summary slide ».

6. Ask for the method whenever there is a derived calculation. If the assistant produces a projection or a ratio, ask it to spell out its assumptions. That is what separates a usable figure from one that needs double-checking.

FAQ and common mistakes

Can the assistant modify my Fincome data?
No. Access is strictly read-only.

Are the figures returned the same as in my dashboard?
Yes. The MCP queries the same calculated metrics, with your calculation options. If a figure differs from what you see on screen, the cause is almost always a period or a scope other than the one you had in mind: ask the assistant to state the dates and filters it used.

Why can't the assistant break figures down by industry or by sales rep?
Because only the product and country breakdowns are native. Every other split relies on your analytical dimensions: if they are not configured in Fincome, the assistant cannot return them.

Can I query my Forecast or my cash position through the MCP?
No. The Forecast, along with cash in, cash out and cash flow variation, is not accessible through the MCP. Those analyses stay in the dashboard.

Do the dates returned match my request?
Date ranges are inclusive by whole period. From 2024-05-01 to 2024-05-31 returns the month of May; from 2024-05-01 to 2024-06-01 returns May and June. Specify your period to remove any ambiguity.

Can I trust a projection produced by the assistant?
The historical data comes from Fincome and is reliable. The projection, however, is a model built by the assistant: always ask it to spell out its assumptions and limitations, and review them as you would any spreadsheet model.

Can I get the list of my raw invoices?
Not yet. The MCP exposes KPIs and refined detailed views (including invoice lines), but not the raw objects from your billing sources. For that, use data exports from Fincome.

Do my colleagues see the same data as me through the MCP?
Everyone connects with their own account and sees what their Fincome permissions allow — no more, no less.

How do I revoke an access?
In the dashboard: Settings → Connected AI assistants.

Related articles

• Connect the Fincome MCP

Updated on: 14/09/2026