# Profitably for AI agents > Profitably's connector gives Claude a governed finance workstation inside Profitably: your data, the metric rules and the tools, so every answer comes straight from your numbers. ## How the connector works The Profitably connector is an MCP server. A model learns it from `help`, then `help `, then `help ` for the exact flags. - **One command tool.** Everything runs through `run_command`, which takes commands like `finmodel forecast --model `. A second tool, `preview_report`, shows report pages as images. - **Help at every level.** `help` lists the command groups, `help finmodel` lists their commands, and `help finmodel set` gives the exact flags. The model learns the whole surface as it goes. - **Predictable replies.** Every command takes named flags, and every reply is a structured result with a machine-readable code, so the model always knows its next step. ## What an AI can do in Profitably - **Read the P&L and every metric:** Starts each conversation with the business's currency, timezone, connected data and recent P&L, then runs read-only SQL over its own analytics: profit by brand, channel, country and product, marketing, cohorts, subscriptions, settlements and banking. Commands: `account bio`, `data query`, `data sample`. - **Explain where a number comes from:** Shows what each table measures, what one row is and exactly how to total it, with live sample rows. Spots gaps worth filling, such as fixed costs to add or an Amazon channel to reconnect. Commands: `schema list`, `schema describe`, `report data-quality`. - **Assign ad spend to brands:** Lists campaigns, ad sets, ads and ad accounts ready to be assigned to a brand, ranked by spend, so every dollar lands on the right brand. Commands: `attribution unlinked`. - **Model scenarios and the financial model:** Reads each brand's real monthly baseline, builds a first 12-month model from the P&L, proposes driver changes such as AOV, nCPA and new orders, copies a model into a separate scenario, quotes the computed forecast and emails it as a branded Excel file. Commands: `finmodel history`, `finmodel create`, `finmodel set`, `finmodel forecast`, `scenario create`, `finmodel export`. - **Plan cash flow:** Quotes the computed cash forecast from the connected bank, from opening balance to the runway low point, and proposes moving, splitting, editing or adding payments. Commands: `cashflow forecast`, `cashflow move`, `cashflow split`, `cashflow add`, `cashflow export`. - **Configure costs:** Reads and bulk-edits every Cost Configuration tab: fixed costs, product costs, shipping rates, per-order rules, gateway fees, order based COGS, variable costs, order exclusions, subscription and influencer costs, and supplier terms. Commands: `costs fixed`, `costs product`, `costs fulfilment`, `costs order-cogs`, `costs import-rows`. - **Build dashboards and alerts:** Builds and edits Custom Dashboards from the business's real metrics, and sets up Custom Metric Alerts after checking the metric's current level and replaying the rule over history. Commands: `dashboard library`, `dashboard create`, `alert preview`, `alert backtest`, `alert create`. - **Produce branded reports:** Renders co-branded PDF reports, slide decks and one-page graphics from a template library, can show the pages in the chat before sending, and follows playbooks for board packs and investor updates. Commands: `report templates`, `report generate`, `report graphic`, `report send`. ## How it keeps answers reliable - **Your data only.** The connection is locked to your business when you approve it. Every query is checked against its BigQuery plan before it runs, so it reads your business's data and yours alone. - **Queries that read, checked first.** Every query reads and leaves your data exactly as it is, and each one is sized and checked before it runs. - **A documented metric catalog.** More than 50 analytics tables are documented with what one row means and exactly how to total it, and the agent reads that guide before its first query. The same rules travel with every answer: one currency per analysis, ratios taken as Profitably computed them, and brand costs attributed directly to the brand that spent them. - **Computed figures, quoted exactly.** Forecast and cash figures come straight from Profitably's own computed forecast, quoted exactly. - **Money changes wait for your approval.** Changes to costs, the financial model and the cash flow forecast arrive as proposals with a before and after preview, and go live the moment you approve them in the chat. - **You stay in control.** Reports are emailed when you confirm them, and you manage the connection any time from your Profitably profile. ## How to connect 1. In Claude, open Settings, then Connectors. 2. Choose Add custom connector. 3. Paste https://mcp.profitably.com/mcp, sign in with your Profitably account and approve the connection. The connector comes with the Profitably AI add-on, free for the first month. It works in Claude on the web, in the desktop app and in Claude Code. The Claude MCP page inside Profitably shows your connector link. The same agent also runs inside Profitably, in the AI panel on every dashboard page, and in Slack, where it answers questions about your numbers. ## Try Profitably Paid plans start with a 14-day free trial, and every trial includes a free report from Profitably's analysts. [Start free trial](https://profitably.com/register) or [compare plans](https://profitably.com/pricing). A guide to the whole site for AI assistants is at https://profitably.com/llms.txt.