mcp (model context protocol) is the standard that lets an llm agent use external tools and data. for developers it means agents that can read codebases; for investors it means something more interesting: your ai assistant can query live deal flow. this is the practical guide.
the one-command setup
npx frontrun-mcp-server --setup
run that in claude code, cursor, codex, or any mcp-capable client. it walks through oauth sign-in (frontrun pro - 7-day free trial) and registers 29 read-only tools over frontrun's investor follow-graph signal.
what your agent can do after
| ask | what happens |
|---|---|
| "what did tracked investors follow this week?" | trending/convergence tools return ranked companies with flag dates |
| "find companies matching 'agent infrastructure'" | semantic thesis search over 12,000+ embedded company descriptions |
| "who founded @engramlab?" | founder resolution with bios and handles |
| "which sub-1k ai companies haven't raised?" | filtered discovery - the exact query behind this list |
| "draft intros to the top 5" | your agent combines the data with your voice and writes the outreach |
the underlying data is the part you can't rebuild with a scraper: continuous follow-graph diffs across 1,000+ tracked investors, with first-flagged provenance. it's how @techdollarhq was on the radar at 13 followers, 123 days before its $3M pre-seed.
mcp vs api: which to use
- mcp - you want an assistant that answers questions and runs sourcing loops interactively. zero code.
- rest api - you're building a product, a scheduled pipeline, or a custom agent harness. 24 endpoints under
/v1; see the api reference.
same data, same auth, same credits (pro includes 10,000/mo).
patterns that work
the morning loop. a scheduled agent run: "diff trending against yesterday, enrich anything net-new in my thesis areas, put a table in slack." the agent version of a daily report, tuned to you.
thesis screening. paste your fund's thesis doc, then: "which tracked companies match, ranked by recent tracked-follow momentum?" semantic search plus live signal in one answer.
meeting prep. "i'm meeting @rialto_xyz tomorrow - what's their signal history?" flag date (jun 5, at 18 followers), momentum since, sector context. thirty seconds.
for the skeptics: why not just scrape x?
x rate-limits, follow lists paginate inconsistently, and the interesting part is the diff over time with provenance - which means running the infrastructure continuously for months before the data means anything. that's the actual product. the protocol layer (mcp) is just how your agent reaches it.
faq
which clients work? claude code, cursor, codex, and anything that speaks mcp. the server is on npm as frontrun-mcp-server.
is there a free way to look first? the trending page is a public slice, no auth.
what does it cost? frontrun pro, $99/mo after a 7-day free trial - includes mcp + api + 10,000 monthly credits.
read next: how to build an ai deal flow agent · best ai deal sourcing tools in 2026