an ai deal flow agent is an llm agent (claude code, cursor, codex, or your own harness) connected to live deal-sourcing data, so one prompt goes from "find companies" to enriched founders and drafted outreach. this guide builds one in about 5 minutes using the frontrun mcp server.
the prompt you will be able to run at the end:
find 10 ai companies under 1k followers that haven't raised - enrich each founder and draft a warm intro from me.
why agents need a signal source (not a database)
most startup databases tell you what already happened: the round was announced, the company registered, the website went live. by then everyone with a pitchbook seat knows.
deal flow agents get interesting when the data is earlier than the announcement. frontrun tracks the follow graphs of 1,000+ venture investors on X and flags companies the moment several of them start following the same small account. real receipts from this signal:
- @techdollarhq - flagged feb 23 at 13 followers. 123 days later: a $3M pre-seed.
- @orthogonal_sh - flagged 184 days before a $4.3M round led by Pantera.
- @rialto_xyz - flagged jun 5 at 18 followers. 26 days before its Robinhood chain launch partner announcement.
- @weaverobotics - flagged ~4.5 months before isaac 1, their home robot, went live.
that is the data your agent will be querying.
step 1: connect the mcp server
in claude code, cursor, or any mcp-capable client:
npx frontrun-mcp-server --setup
the setup walks you through oauth sign-in (frontrun pro, 7-day free trial). once authed, your agent has 29 read-only tools over the live signal: trending companies, semantic thesis search, company lookups, founder resolution, convergence queries, and more.
works with claude code, cursor, codex, and any client that speaks mcp.
step 2: give the agent its loop
a deal flow agent has three stages. you can run them as one prompt or wire them into a scheduled workflow:
- discover - "what are tracked investors following this week?" (trending / convergence tools), or thesis-first: "companies matching 'agent infrastructure for regulated industries'" (semantic search over 12,000+ embedded companies).
- enrich - resolve the founders behind each company, pull bios, follower quality, what the company shipped recently.
- act - draft the outreach in your voice, output a table or csv, push to your pipeline. you review and send.
example prompts that work out of the box:
find 10 ai companies under 1k followers that haven't raised -
enrich each founder and draft a warm intro from me.
which companies gained 3+ tracked follows in the last 7 days?
sector-tag them and rank by follower count ascending.
search my thesis: "stablecoin infrastructure for emerging markets."
return the 5 best matches with founders and one-line pitches.
step 3: schedule it
the difference between a demo and deal flow is cadence. run the discover step daily (a cron hitting your agent, or claude code in a scheduled job), diff against yesterday, and only surface net-new companies. that mirrors how frontrun's own daily reports work - net-new only, no repeats.
step 4 (optional): go direct to the api
if you outgrow mcp or want this in your own product, the same data is a rest api - 24 endpoints under /v1, including GET /v1/search/thesis for semantic matching. see the api reference and authentication.
faq
do i need to write code? no. connect the mcp server to claude code or cursor and prompt in plain english.
what does it cost? frontrun pro is $99/mo with a 7-day free trial - that includes api + mcp access and 10,000 monthly credits.
can it send dms for me? drafting is the agent's job; sending stays yours. human-in-the-loop outreach converts better and keeps your account safe.
where does the signal come from? the follow graphs of 1,000+ tracked investors, polled continuously, with first-flagged provenance on every company. counts are public on the trending page - which investors are behind them is in the product.
see the signal live: trending startups today · read next: the best ai deal sourcing tools in 2026