I'm Asking Mattermark's Question Again

In 2013 I started Mattermark to answer one question: which startups are growing? The person asking sat at a browser, so we built a dashboard.

StartupGraph asks the same question again. This time the one asking is an agent, so there is no dashboard. It is an API and an MCP server, and it is open as a beta at startupgraph.dev.

Spreadsheets that needed to be code

Mattermark started on this blog. After we shut down Referly, I spent the spring of 2013 ranking startups in posts: Alexa rank, Twitter followers, Facebook likes, App Store rank. One list of a hundred companies took me twenty hours to collect, and as soon as I had collected it, it was stale again. These spreadsheets needed to be code, so we wrote crawlers and launched Mattermark six weeks later.

The posts also brought corrections. When I used Crunchbase data to list the venture firms that looked inactive, hundreds of calls and emails came back telling me what I had wrong. That was how we found out how little good data about startups was public.

A screen, saved

By 2017 I was asking the question as a screen. For a post naming Bay Area startups ripe for a Series A, I filtered Mattermark for companies near fifteen people, whose team had grown over the last six months, nine to twenty-four months after their last round. Saved, the search emailed me whenever a new company matched.

FullContact bought Mattermark in December 2017. The Mattermark Daily sent its last issue the next May.

What is different now

You ask Claude which companies in your funds added people this quarter, and Claude asks StartupGraph. A screen is something your agent saves and runs again. A portfolio records a company's headcount, total raised, and last round when you add it, and shows what changed since. A roll-up follows a fund's deals. A feed lists every fact added or changed since a date.

The corrections changed the most. An agent can submit a fact along with the page it came from. The fact goes live marked unverified. StartupGraph fetches that page, looks for the company and the figure, and only then marks it verified. A verified fact earns the agent's key more calls. A source that does not exist is a strike against the key, and three strikes revoke it.

Mattermark had seventy people at its peak. Most of StartupGraph went in on one day, October 12, built with Claude.

What does not work yet

The question I most want answered is who raised in the last ninety days and is hiring. It is the 2017 screen with a shorter window, and StartupGraph cannot answer it yet. Rounds after February are thin, and open roles have not had their first weekly pull. Headcount histories exist, but they are not backed by a source, so they are served as unverified.

The coverage endpoint tells your agent what is missing, so it can say "I don't know" instead of guessing. The roadmap is public and says what is true today. I'll write when things ship, and when they slip.

Getting a key

Keys go out by email, in order, and you get your number the moment you sign up at startupgraph.dev. If you have a source it should be pulling from, tell your agent to submit it, or write to me.

What works today, how a fact gets checked, what the data does not have yet, and how Mattermark turned into this are all on the StartupGraph page.