Seven People Built the Thing That Lets AI Assistants Read Your Live Business Data
A seven-person bootstrapped company just published a URL that lets any AI assistant read your live business data: the kind of live-database integration that usually takes enterprises months to build.
Basedash, a Y Combinator Summer 2020 company, published the endpoint when it launched its Dashboard Agent on Product Hunt on April 30, 2026. The URL — https://charts.basedash.com/api/public/mcp — connects any compatible AI client to the databases and SaaS tools wired into a Basedash workspace, so an assistant like Claude Code or Cursor can answer business questions against live data instead of static documents. According to Basedash's MCP server documentation, the two tools the server exposes are asking business questions in plain English and listing which data sources are connected. Every query respects the permissions already configured in Basedash, so the AI inherits whatever row-level and column-level access a human analyst would have. The authentication layer uses OAuth: adding the server URL to an AI client triggers a browser flow that ties the session to the user's Basedash account, passing the existing permission graph without separate credentials or API keys. The Dashboard Agent built on top of it has a 5.0 rating from early users.
What makes this worth noting is not the dashboard product. Basedash employed seven people, was bootstrapped, and had roughly $1 million in annual recurring revenue as of 2024, according to a competitive comparison published this week by rival Sequel. That same comparison says Basedash holds SOC 2 Type II certification and that customer data is never used to train its AI models. The company started in 2020 as a YC S20 batch project before pivoting in 2024 from a developer database GUI into an AI-native business intelligence platform.
The Sequel comparison also shows the two architectural bets currently competing in AI business intelligence. Sequel's agents accumulate context over time — they learn a business's schema, terminology, and query patterns across sessions, so questions asked in week eight carry everything the system observed in weeks one through seven. Basedash's dashboard agent is transactional: describe what you want, it builds the chart. Whether that is a limitation or a feature depends on whether you trust a self-learning system to improve on its own or want to specify exactly what you are building.
For the agent infrastructure ecosystem, the more interesting question is what a small team shipping a working MCP-connected AI product tells the market about tooling maturity. The standard narrative around enterprise AI adoption holds that wiring a model to a live data source requires significant integration work: authentication, permission translation, query scoping, output validation. Basedash appears to have compressed that into a documented server any MCP-compatible client can use — a proof point, if it holds under messy real-world conditions, that the gap between prompting a model and shipping something an enterprise will actually use is narrower than the market assumes.
The $250 per month entry point for two users, with no free tier, prices Basedash for teams that have already decided AI BI is worth a dedicated budget line. The 14-day trial with full access and no credit card required is the evaluation hook. What comes after is a $3,000 annual commitment before a third person can log in, which filters out smaller teams that have not committed to an annual contract. Growth tier runs $1,000 per month for 25 users and unlocks the full library of 750-plus data source integrations.
What to watch next is whether the MCP-connected workflow survives contact with the messy reality of enterprise data: schemas that do not fit cleanly into Basedash's permission model, queries that need joins across sources the system has not seen before, permission boundaries that break down at the edges of what the access control graph can represent. The Product Hunt rating is from early users who knew exactly what they were evaluating. The harder test is the team that spins it up hoping to retire the BI vendor they have been meaning to cancel.