Prompts to paste into Claude Code once the MCP server is connected. Each one has the agent build something on your machine with StartupGraph data. Change names and criteria to yours.
claude mcp add --transport http --scope user startupgraph https://startupgraph.dev/mcp --header "Authorization: Bearer $STARTUPGRAPH_KEY"
You are an LP in a few funds and want one page, on your machine, that shows their portfolio companies and what moved.
Using the StartupGraph MCP server, create a fund roll-up called "LP book" for Sequoia Capital and Andreessen Horowitz. Then build a small local web app (one HTML page served by a Node script) that reads the roll-up when I open it, lists every company with headcount, last round and status, and puts anything that changed since my last visit at the top. Read my key from the STARTUPGRAPH_KEY environment variable and never write it to disk.
You have a search you keep repeating. Save it once and wrap it in a command.
Save a StartupGraph screen named "AI over 100" for category ai_ml with min_headcount 100, sorted by current_headcount descending. Write a CLI called sg-screen that runs any saved screen by name and prints a table with slug, headcount, last round and total raised. Add --csv to write the same rows to a file.
You track a set of companies and want to hear only when something changes.
Create a StartupGraph portfolio called "Watching" with Anthropic, Cursor, Ramp and Vercel. Write a script I can run from cron every Monday that calls changes_since for the past seven days scoped to that portfolio, and writes a plain-text summary to ~/startupgraph/weekly.txt: one line per change, with the source URL. If nothing changed, write that.
Open roles come from public job boards and start pulling weekly. Build the view now and it fills in as the data lands.
Run my saved StartupGraph screen "AI over 100" and, for each company, call open_jobs. Build a local page that ranks the companies by open roles and shows the change since last week. Mark companies where no job board has been found yet instead of showing zero.
An agent that knows the gaps gives better answers. Make it ask first.
Before answering any question about startups using StartupGraph, call coverage for the fields and countries the question touches. If coverage is thin, say what you do not know before giving the answer, and cite the source URL on every fact you use.
You read funding news. Let the agent file it with a source and earn calls on your key.
Here is a funding announcement: <paste URL>. Read it, extract the company, round type, amount and date, and submit it to StartupGraph with submit_fact using that URL as the source. If the company is not in StartupGraph, submit_company first. Then check fact_status and tell me whether it verified.
See who a firm invests alongside, from round data.
Use search_investors to find Sequoia Capital on StartupGraph, then funding_rounds filtered to that investor. Count which other investors appear in the same rounds and build a local page with a table of co-investors ranked by shared deals, each row linking to the companies involved.
You live in a spreadsheet. Keep it fed.
Write a script that runs my StartupGraph screen "Fintech seed" and writes the result to a Google Sheet (or a local CSV if I have no credentials), one row per company, replacing the sheet contents each run. Add a column with the date each row last changed, using changes_since.