SaaSFlow for AI agents

The finance data of a SaaS company, readable by Claude, ChatGPT, Cursor, or the agent you build yourself. One MCP endpoint, a public REST API, and a CLI, all on the same numbers your dashboard shows.

Connect an assistant Build your own I'm an agent

Assistants

Connect an assistant in a minute

The MCP server lives at mcp.saasflow.com and speaks OAuth, so your assistant needs nothing more than the address. Pick yours:

1
Open the connector settings in Claude.
Open Claude connectors
2
Click "Add custom connector", name it SaaSFlow and paste this address:
https://mcp.saasflow.com
3
Click Connect and sign in. The connector works in Claude on the web, in the desktop and mobile apps, and in Code sessions of the desktop app.

Sign in or create your account

When you connect SaaSFlow, a SaaSFlow page opens in your browser. Sign in, or create a free account right there, and approve the connection.

Reading your data is always included. "Make changes to company data" lets the agent do the setup work, and "Create companies for you" sets up your company if you don't have one yet. A new account gets every option ticked; you can untick what you don't want. The agent only sees the tools of the options you allow, so to add one later, disconnect and connect again.

Paste this into a new chat

A new chat or session picks up the SaaSFlow tools you just connected.

I've connected SaaSFlow. Set it up for my company. Instructions for agents: https://saasflow.com/agents.md

You connect Stripe and your bank
The agent sends you a link into the app. You sign in with each provider or upload its export, so this part stays with you.
Your agent does the tedious part
It categorizes hundreds of transactions, creates vendors and customers, adds planned income and costs to the forecast, and shows you each change before applying it.

Want MRR and cash in the menu bar instead? Get the Raycast extension →

Tools

What an agent can ask once connected

Over 90 read tools cover companies, accounts, transactions, customers, vendors, subscriptions, and every metric the dashboards show. The server tells the model to take metrics from the data tools instead of summing transactions itself, so answers match the app.

“What is our MRR today, and how did it move this month?”: data_mrrAtDate
“Which customers churned or downgraded in March?”: data_customerChanges
“Show me last quarter's P&L split by category.”: data_profitLossData
“Who are our ten biggest expense vendors this year?”: data_topExpenseVendors
“How long does the cash last at the current burn?”: data_cashFlowTimeSeries
“Find the Stripe payout from January 12 and show its transactions.”: search
“How does the 2025 cohort retain compared to 2024?”: data_cohorts
“What exactly is counted in net revenue retention here?”: data_metricExplanation

Full tool reference →

Developers

Build your own agent

Everything the MCP tools do is a REST call underneath. Use the API directly, script it with the CLI, or ship an app that users install with OAuth.

REST API

JSON over HTTPS, resources scoped by company, and one data endpoint per metric slice. A batch endpoint runs several slices in one round trip.

curl

curl https://api.saasflow.com/companies/cmp_.../data/mrrAtDate \
  -H "Authorization: Bearer sf_live_..." \
  --get --data-urlencode "mrrType=cmrr" \
  --data-urlencode "date=2026-08-31"

API overview →

CLI

The saasflow package on npm signs in with OAuth or an API key, prints JSON for pipes and tables for people, and has one subcommand per data slice.

Terminal

npm install -g saasflow
saasflow login
saasflow data mrrAtDate --mrr-type cmrr --date 2026-08-31
saasflow transactions list --limit 20 --format table

CLI docs →

Typed clients and OAuth apps

Generate types from the OpenAPI document in one line. For an app other people install, register an OAuth client and request the read or write scope.

Terminal

npx openapi-typescript https://api.saasflow.com/openapi.json -o saasflow.d.ts

OAuth for apps →

In the product

Agents that already run inside SaaSFlow

You do not have to bring your own model to get value from agents. Three run inside SaaSFlow today, and each can also be started through the API.

AI analyst

Pick a closed month and an agent works through MRR movements, customer changes, spending, and cash. It writes a report with its own layout and charts, and the headline numbers are replayed against live data before you see it.

Report chat

Ask where a number comes from, or have the analyst cut a section, add a breakdown, or start from an earlier version. Every accepted edit passes the same checks and becomes a new version.

Categorization helper

Reviews uncategorized transactions and proposes vendors, customers, and categories. It never writes on its own: you review the suggestions and apply the ones you agree with.

Each run executes in an isolated container with a credential that only lasts for that run. The models run on Google Cloud in the EU, the same provider that hosts SaaSFlow, and no other AI provider receives your data. More on security

Discovery

If you are an agent, start here

Every document a machine needs is at a stable, unauthenticated URL. Fetch these before guessing.

saasflow.com/llms.txt: What SaaSFlow is, when to use it, and where every page lives.
saasflow.com/agents.md: This page as Markdown, starting with the step-by-step guide for an agent that was asked to set up SaaSFlow.
docs.saasflow.com/llms.txt: Page-by-page index of the documentation. Append .md to any docs URL for its Markdown source.
api.saasflow.com/openapi.json: The OpenAPI 3.1 document, with the permission each operation needs.
api.saasflow.com/.well-known/mcp: MCP server card: the endpoint and what it accepts for auth.
api.saasflow.com/.well-known/api-catalog: RFC 9727 API catalog pointing at the same places.
document.modelContext: Browser agents on this site can call the WebMCP tools estimate_saasflow_price and estimate_saas_valuation instead of reading the pricing and valuation pages.

Permissions

What an agent can and cannot do

Scoped credentials: An API key belongs to one company and carries an explicit permission set. An OAuth grant is limited to the scopes you approve, intersected with your own role.
You choose what it may change: Reading is always included. Making changes, managing settings and creating companies are separate options when you connect, and an agent only sees the tools of the options you allowed.
Admin access only when you grant it: Members and API keys stay with signed-in people. An agent can change company settings and manage integrations only if you tick that when you connect it, or give its API key those permissions.
Revocation and audit: Revoking a key or a grant in Settings takes effect on the next request. Every tool call over the hosted endpoint is written to an audit log with the credential that made it.
Sandboxed built-in agents: The agents that run inside SaaSFlow execute in isolated containers with a run-scoped credential. They run on Claude, hosted by Google Cloud in the EU.
EU hosted, no money movement: SaaSFlow runs in the EU and reads from your billing and banking providers. Nothing here can initiate a payment.