this is the reference page. what the server exposes, grouped by what you would actually ask an agent to do.
install
npx frontrun-mcp-server --setup
writes the config for claude code, cursor, codex or any mcp-capable client. the oauth wall appears at first tool call, which is where a trial starts.
for a remote connection instead:
https://www.frontrun.vc/mcp
Authorization: Bearer <api key>
streamable http. keys are in settings. mcp and rest api access are on the pro plan at $99/mo.
the 37 tools
tracking, what you are watching
frontrun_status frontrun_list_tracked frontrun_track frontrun_untrack frontrun_preview
add or remove investor accounts from your tracked set and check quota. preview shows what an account would surface before you spend a slot on it.
signal, the part nothing else has
frontrun_new_follows frontrun_snapshot frontrun_enriched_follows frontrun_convergence frontrun_trending frontrun_feed frontrun_discover
new_follows answers who did tracked investors start following in this window. convergence answers which accounts had several of them arrive at once, which is the actual buy signal. enriched_follows returns those with classification and context attached so the agent is not reasoning over bare handles.
search and classification
frontrun_search frontrun_thesis_search frontrun_classify frontrun_sectors
thesis_search takes natural language, so an agent can pass a fund's thesis straight through: "pre-seed robotics companies where the founder came out of an autonomy team".
company and founder data
frontrun_company frontrun_company_founders frontrun_founders_batch frontrun_company_signals frontrun_company_resources frontrun_company_funding
founders_batch matters more than it looks. an agent evaluating thirty companies makes one call instead of thirty, which is the difference between a workflow that runs and one that times out.
investor analysis
frontrun_vc_activity frontrun_vc_similar frontrun_warm_path
warm_path answers the question that decides whether outreach happens: who do we both know.
workflow and output
frontrun_create_rule frontrun_list_rules frontrun_update_rule frontrun_delete_rule frontrun_tag frontrun_list_tags frontrun_reports frontrun_list_webhooks frontrun_create_webhook frontrun_delete_webhook frontrun_draft_dm frontrun_send_dm
rules and webhooks let the agent set up its own standing alerts rather than being re-prompted. draft_dm and send_dm close the loop.
what this gives an agent that web search does not
an agent with web search can read anything already published. the three things it cannot get that way:
timing. who started following whom, and when. that is a diff, and it exists only because something was recording. no crawler recovers it later.
a judged verdict. raw signal is mostly noise. handing an agent 2,549 unfiltered rows produces a worse answer than handing it the 896 that survived a career check, because the model will rank the noise with total confidence.
structure. typed tools returning typed data, instead of an agent parsing html and hallucinating a founder name from a footer.
a prompt that uses several at once
surface accounts with 3+ new tracked-investor follows in the last 7 days. for each, pull the founders and the funding history. drop anything that has already announced a round. rank the rest against this thesis: early robotics and applied autonomy, technical founders, pre-seed. for the top three, check whether we have a warm path, and draft a first message for the one where we do.
that is convergence, founders_batch, company_funding, thesis_search, warm_path and draft_dm, chosen and sequenced by the agent.
related
open source deal flow
companies and founders surfaced on frontrun. curated, every week.
