> ## Knowledge Base Index > Fetch the complete knowledge base index at: https://help.fincome.co/sitemap.xml > Use this file to discover available pages before exploring further. > Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash). # Connect the Fincome MCP Fincome's MCP server connects an AI assistant (Claude, Cursor…) to your financial KPIs. You ask your questions in natural language, and the agent fetches the figures directly from Fincome — no exports, no copy-pasting. # What is Fincome's MCP server? **MCP** (Model Context Protocol) is an open standard that lets an AI assistant connect securely to a data source. Concretely, Fincome's MCP server gives an agent like **Claude** or **Cursor** **read-only** access to your financial **KPIs**: MRR, ARR, churn, revenue, LTV/CAC, etc. | Note — The MCP server gives access to your aggregated KPIs as well as refined detailed views (e.g. ARR/MRR by customer and by month, ARR movements, billing lines). It does not yet give access to the raw objects of your billing sources (invoices, customers, and subscriptions as imported) — this will come in a future version. The agent connects **as one of your users** and sees exactly the same data that this user can view in the dashboard — no more, no less. # What is it useful for, concretely? The MCP server is useful whenever you want to **get a figure or an analysis without going through the interface**. A few typical use cases: * **Answering a one-off question** without building a report: *"How many new customers in May?"*, *"What is my churn rate this quarter?"* * **Preparing an investor meeting or a board**: ask the agent to retrieve MRR, ARR, growth, and churn over the period, then format them. * **Breaking down a KPI by dimension**: *"What is my ARR by product?"*, *"My churn by country this quarter."* * **Crossing your analyses with other sources** already connected to your assistant, in a single conversation thread. The benefit: your Fincome figures become accessible **where you already work** with AI, in natural language. | To discover everything you can produce with the MCP (due diligence datapack, 36-month ARR projections, on-the-fly answers), see [What can you do with the Fincome MCP? Use cases and example analyses](https://help.fincome.co/en/article/what-can-you-do-with-the-fincome-mcp-use-cases-and-example-analyses-1ii92jh/). # How to install and test it (beta) || The MCP server is currently in beta. It works with **any MCP-compatible assistant** (Claude, Cursor, Claude Desktop, and other clients of the protocol). In all cases, the connection uses the same credentials as your Fincome dashboard. ### Step 1 — Add the MCP server to your assistant The way to add an MCP server depends on your assistant, but the principle is always the same: you declare a connector pointing to the Fincome server URL. ``` https://mcp.fincome.co/mcp ``` * **Assistants with a connectors interface**: open the connectors section, choose "add a custom connector", then paste the URL above. * **Assistants configured by file** (Cursor, Claude Desktop, etc.): declare the server in your `mcp.json` file. See the [MCP server's technical documentation](https://docs.fincome.co/mcp). || Using Claude? The simplest way: Fincome is available in **Claude's connector directory**. Search for "Fincome" in the directory and click **Connect** — it's added in one click, with no URL to enter. ### Step 2 — Enable it in your conversation Once added, **enable the Fincome connector** in the conversation where you want to use it. It then becomes available to the assistant. ### Step 3 — Connect your Fincome account On first use, your assistant redirects you to the **Fincome login page** (same credentials as the dashboard). You **approve the access**, and the agent receives a secure token, limited to your user and your company. That's it: you can ask your questions. || Good to know — The connection is done via OAuth 2.1: no password is shared with the assistant, and access can be revoked at any time from Settings → Connected AI assistants in the dashboard. # Managing several entities? Each connection is linked to **a single account** (one user and one company). If you manage **several separate entities**, simply point each account to a **different URL**, for example: ``` https://mcp.fincome.co/mcp2 https://mcp.fincome.co/mcp3 ``` Add one connection per entity this way, and authorize each one with the corresponding account. || The URL number (/mcp2, /mcp3…) does not identify a specific entity: it is just a technical "slot" to work around refusals of the same URL twice. The entity is chosen at connection time. # What about Cursor or Claude Desktop? It is also possible to configure it via an `mcp.json` file. For these cases, refer directly to the [MCP server's technical documentation](https://docs.fincome.co/mcp). # What you can ask the agent The assistant automatically queries your KPIs based on your question. You have nothing to configure — just phrase your request in natural language. | You want… | Example question | | ---- | | List the available metrics | *"Which Fincome metrics can I view?"* | | A time series of a KPI | *"Show me my monthly MRR over the last 12 months."* | | A KPI broken down by dimension | *"What is my ARR by product?"*, *"My churn by country this quarter."* | | A combined analysis | *"Compare my MRR for January and June, and explain the gap."* | Behind the scenes, the agent retrieves your metrics (MRR, ARR, churn, revenue, LTV/CAC…), filters them, breaks them down by dimension (including your **custom dimensions**), and chooses the time step (month, quarter, year). # Security and privacy Security is at the heart of how the MCP server works: * **Read-only**: the agent can view your data, but **cannot modify anything** in Fincome. * **Your permissions apply**: the agent sees exactly what the connected user sees, according to their rights in the dashboard. * **OAuth 2.1 authentication**: no password is transmitted to the assistant; access relies on a secure token, limited to your user and your company. * **Revocable at any time**: from **Settings → Connected AI assistants**, you can view and delete each connection. These guarantees add to the platform's compliance (SOC 2, GDPR, ISO 27001, European hosting). --- # What we produced in 10 minutes To give you a concrete idea, here is a **financial review datapack** (due diligence style) built end to end via the Fincome MCP — on a demo account, in about ten minutes, **without any export or SQL**. Each time, it was enough to ask the question in natural language and ask the assistant to format the result: * **ARR breakdown** by industry, account size, acquisition channel, and product. * **Recurring vs one-off** and level of deferred revenue. * **Collection** by status and unpaid rate. * **Average basket (ARPA)** and **LTV** over 12 rolling months, overall and by segment. * **Growth breakdown** (new / reactivation / expansion / contraction / churn). * **Acquisition by sales rep**, **upsell & NRR by CSM**, and **24-month NRR cohorts**. * A **summary scorecard** of SaaS KPIs (ARR, YoY growth, NRR, GRR, quick ratio, churn, recurring share…). Everything was assembled directly into a spreadsheet and a presentation from the agent's answers. 1. [Presentation](https://26202015.fs1.hubspotusercontent-eu1.net/hubfs/26202015/MCP/Fincome%20-%20Draft%20analyses%20v1%20MCP%20SaaSCo%20-%20juin-26.pdf) 2. [Excel file](https://26202015.fs1.hubspotusercontent-eu1.net/hubfs/26202015/MCP/01%20Fincome%20-%20Draft%20analyses%20v1%20MCP%20SaaSCo%20-%20juin-26.xlsx) || Want to see more deliverables? The article [What can you do with the Fincome MCP? Use cases and example analyses](https://help.fincome.co/en/article/what-can-you-do-with-the-fincome-mcp-use-cases-and-example-analyses-1ii92jh/) details three complete examples (due diligence datapack, 36-month ARR projections, and on-the-fly answers). --- # FAQ and common errors **Which assistants are compatible?** All clients that support the MCP protocol, such as **Claude** (app and Desktop) and **Cursor**. The connection is made either via the connector directory (Claude), or via an `mcp.json` file (Cursor, Claude Desktop). **Can the agent modify my data?** No. Access is **strictly read-only**. The assistant can query your metrics and your views, but cannot create, modify, or delete anything in Fincome. **Does the agent see all my company's data?** It sees exactly what **the connected user** sees, according to their Fincome permissions. If a user does not have access to certain data in the dashboard, the agent will not have access to it either. **How to connect several companies at the same time?** Each connection is linked to one user and one company. To connect several, assign each one **its own URL** (`…/mcp`, `…/mcp2`, … up to `…/mcp10`) and authorize each account separately. **How to revoke an access?** In the dashboard, go to **Settings → Connected AI assistants**. You will find all the active connections there and can delete them with one click. **Are my credentials shared with the assistant?** No. The connection goes through **OAuth 2.1**: you authenticate on the Fincome page, and the assistant only receives a secure, revocable token, never your password. **Do the returned dates really match my request?** Date ranges are **inclusive by whole period**. For example, 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 in your question to remove any ambiguity.